MyArxiv
Cryptography and Security 27
DynaMark: A Reinforcement Learning Framework for Dynamic Watermarking in Industrial Machine Tool Controllers
Industry 4.0's highly networked Machine Tool Controllers (MTCs) are prime targets for replay attacks that use outdated sensor data to manipulate actuators. Dynamic watermarking can reveal such tampering, but current schemes assume linear-Gaussian dynamics and use constant watermark statistics, making them vulnerable to the time-varying, partly proprietary behavior of MTCs. We close this gap with DynaMark, a reinforcement learning framework that models dynamic watermarking as a Markov decision process (MDP). It learns an adaptive policy online that dynamically adapts the covariance of a zero-mean Gaussian watermark using available measurements and detector feedback, without needing system knowledge. DynaMark maximizes a unique reward function balancing control performance, energy consumption, and detection confidence dynamically. We develop a Bayesian belief updating mechanism for real-time detection confidence in linear systems. This approach, independent of specific system assumptions, underpins the MDP for systems with linear dynamics. On a Siemens Sinumerik 828D controller digital twin, DynaMark achieves a reduction in watermark energy by 70% while preserving the nominal trajectory, compared to constant variance baselines. It also maintains an average detection delay equivalent to one sampling interval. A physical stepper-motor testbed validates these findings, rapidly triggering alarms with less control performance decline and exceeding existing benchmarks.
OptMark: Robust Multi-bit Diffusion Watermarking via Inference Time Optimization
Watermarking diffusion-generated images is crucial for copyright protection and user tracking. However, current diffusion watermarking methods face significant limitations: zero-bit watermarking systems lack the capacity for large-scale user tracking, while multi-bit methods are highly sensitive to certain image transformations or generative attacks, resulting in a lack of comprehensive robustness. In this paper, we propose OptMark, an optimization-based approach that embeds a robust multi-bit watermark into the intermediate latents of the diffusion denoising process. OptMark strategically inserts a structural watermark early to resist generative attacks and a detail watermark late to withstand image transformations, with tailored regularization terms to preserve image quality and ensure imperceptibility. To address the challenge of memory consumption growing linearly with the number of denoising steps during optimization, OptMark incorporates adjoint gradient methods, reducing memory usage from O(N) to O(1). Experimental results demonstrate that OptMark achieves invisible multi-bit watermarking while ensuring robust resilience against valuemetric transformations, geometric transformations, editing, and regeneration attacks.
Entropy-Based Non-Invasive Reliability Monitoring of Convolutional Neural Networks
Convolutional Neural Networks (CNNs) have become the foundation of modern computer vision, achieving unprecedented accuracy across diverse image recognition tasks. While these networks excel on in-distribution data, they remain vulnerable to adversarial perturbations imperceptible input modifications that cause misclassification with high confidence. However, existing detection methods either require expensive retraining, modify network architecture, or degrade performance on clean inputs. Here we show that adversarial perturbations create immediate, detectable entropy signatures in CNN activations that can be monitored without any model modification. Using parallel entropy monitoring on VGG-16, we demonstrate that adversarial inputs consistently shift activation entropy by 7% in early convolutional layers, enabling 90% detection accuracy with false positives and false negative rates below 20%. The complete separation between clean and adversarial entropy distributions reveals that CNNs inherently encode distribution shifts in their activation patterns. This work establishes that CNN reliability can be assessed through activation entropy alone, enabling practical deployment of self-diagnostic vision systems that detect adversarial inputs in real-time without compromising original model performance.
comment: 8 pages, 3 figures, 2 tables
Cybersecurity AI: Hacking the AI Hackers via Prompt Injection
We demonstrate how AI-powered cybersecurity tools can be turned against themselves through prompt injection attacks. Prompt injection is reminiscent of cross-site scripting (XSS): malicious text is hidden within seemingly trusted content, and when the system processes it, that text is transformed into unintended instructions. When AI agents designed to find and exploit vulnerabilities interact with malicious web servers, carefully crafted reponses can hijack their execution flow, potentially granting attackers system access. We present proof-of-concept exploits against the Cybersecurity AI (CAI) framework and its CLI tool, and detail our mitigations against such attacks in a multi-layered defense implementation. Our findings indicate that prompt injection is a recurring and systemic issue in LLM-based architectures, one that will require dedicated work to address, much as the security community has had to do with XSS in traditional web applications.
I Stolenly Swear That I Am Up to (No) Good: Design and Evaluation of Model Stealing Attacks
Model stealing attacks endanger the confidentiality of machine learning models offered as a service. Although these models are kept secret, a malicious party can query a model to label data samples and train their own substitute model, violating intellectual property. While novel attacks in the field are continually being published, their design and evaluations are not standardised, making it challenging to compare prior works and assess progress in the field. This paper is the first to address this gap by providing recommendations for designing and evaluating model stealing attacks. To this end, we study the largest group of attacks that rely on training a substitute model -- those attacking image classification models. We propose the first comprehensive threat model and develop a framework for attack comparison. Further, we analyse attack setups from related works to understand which tasks and models have been studied the most. Based on our findings, we present best practices for attack development before, during, and beyond experiments and derive an extensive list of open research questions regarding the evaluation of model stealing attacks. Our findings and recommendations also transfer to other problem domains, hence establishing the first generic evaluation methodology for model stealing attacks.
comment: Under review
Analogy between Learning With Error Problem and Ill-Posed Inverse Problems
In this work, we unveil an analogy between well-known lattice based learning with error problem and ill-posed inverse problems. We show that LWE problem is a structured inverse problem. Further, we propose a symmetric encryption scheme based on ill-posed problems and thoroughly discuss its security. Finally, we propose a public key encryption scheme based on our symmetric encryption scheme and CRYSTALS-Kyber KEM (key encapsulation mechanism) and discuss its security.
Detecting Stealthy Data Poisoning Attacks in AI Code Generators
Deep learning (DL) models for natural language-to-code generation have become integral to modern software development pipelines. However, their heavy reliance on large amounts of data, often collected from unsanitized online sources, exposes them to data poisoning attacks, where adversaries inject malicious samples to subtly bias model behavior. Recent targeted attacks silently replace secure code with semantically equivalent but vulnerable implementations without relying on explicit triggers to launch the attack, making it especially hard for detection methods to distinguish clean from poisoned samples. We present a systematic study on the effectiveness of existing poisoning detection methods under this stealthy threat model. Specifically, we perform targeted poisoning on three DL models (CodeBERT, CodeT5+, AST-T5), and evaluate spectral signatures analysis, activation clustering, and static analysis as defenses. Our results show that all methods struggle to detect triggerless poisoning, with representation-based approaches failing to isolate poisoned samples and static analysis suffering false positives and false negatives, highlighting the need for more robust, trigger-independent defenses for AI-assisted code generation.
comment: Accepted to the 3rd IEEE International Workshop on Reliable and Secure AI for Software Engineering (ReSAISE, 2025), co-located with ISSRE 2025
Hybrid Cryptographic Monitoring System for Side-Channel Attack Detection on PYNQ SoCs SC'25
AES-128 encryption is theoretically secure but vulnerable in practical deployments due to timing and fault injection attacks on embedded systems. This work presents a lightweight dual-detection framework combining statistical thresholding and machine learning (ML) for real-time anomaly detection. By simulating anomalies via delays and ciphertext corruption, we collect timing and data features to evaluate two strategies: (1) a statistical threshold method based on execution time and (2) a Random Forest classifier trained on block-level anomalies. Implemented on CPU and FPGA (PYNQ-Z1), our results show that the ML approach outperforms static thresholds in accuracy, while maintaining real-time feasibility on embedded platforms. The framework operates without modifying AES internals or relying on hardware performance counters. This makes it especially suitable for low-power, resource-constrained systems where detection accuracy and computational efficiency must be balanced.
comment: This paper is submitted at Supercomputing (SC'25)
Condense to Conduct and Conduct to Condense
In this paper we give the first examples of low-conductance permutations. The notion of conductance of permutations was introduced in the paper "Indifferentiability of Confusion-Diffusion Networks" by Dodis et al., where the search for low-conductance permutations was initiated and motivated. In this paper we not only give the desired examples, but also make a general characterization of the problem -- i.e. we show that low-conductance permutations are equivalent to permutations that have the information-theoretic properties of the so-called Multi-Source-Somewhere-Condensers.
Agentic Discovery and Validation of Android App Vulnerabilities
Existing Android vulnerability detection tools overwhelm teams with thousands of low-signal warnings yet uncover few true positives. Analysts spend days triaging these results, creating a bottleneck in the security pipeline. Meanwhile, genuinely exploitable vulnerabilities often slip through, leaving opportunities open to malicious counterparts. We introduce A2, a system that mirrors how security experts analyze and validate Android vulnerabilities through two complementary phases: (i) Agentic Vulnerability Discovery, which reasons about application security by combining semantic understanding with traditional security tools; and (ii) Agentic Vulnerability Validation, which systematically validates vulnerabilities across Android's multi-modal attack surface-UI interactions, inter-component communication, file system operations, and cryptographic computations. On the Ghera benchmark (n=60), A2 achieves 78.3% coverage, surpassing state-of-the-art analyzers (e.g., APKHunt 30.0%). Rather than overwhelming analysts with thousands of warnings, A2 distills results into 82 speculative vulnerability findings, including 47 Ghera cases and 28 additional true positives. Crucially, A2 then generates working Proof-of-Concepts (PoCs) for 51 of these speculative findings, transforming them into validated vulnerability findings that provide direct, self-confirming evidence of exploitability. In real-world evaluation on 169 production APKs, A2 uncovers 104 true-positive zero-day vulnerabilities. Among these, 57 (54.8%) are self-validated with automatically generated PoCs, including a medium-severity vulnerability in a widely used application with over 10 million installs.
Generalized Encrypted Traffic Classification Using Inter-Flow Signals
In this paper, we present a novel encrypted traffic classification model that operates directly on raw PCAP data without requiring prior assumptions about traffic type. Unlike existing methods, it is generalizable across multiple classification tasks and leverages inter-flow signals - an innovative representation that captures temporal correlations and packet volume distributions across flows. Experimental results show that our model outperforms well-established methods in nearly every classification task and across most datasets, achieving up to 99% accuracy in some cases, demonstrating its robustness and adaptability.
comment: Accepted manuscript at Availability, Reliability and Security (ARES 2025), published in Lecture Notes in Computer Science, vol. 15992, Springer, Cham. DOI: https://doi.org/10.1007/978-3-032-00624-0_11
Towards a Decentralized IoT Onboarding for Smart Homes Using Consortium Blockchain
The increasing adoption of smart home devices and IoT-based security systems presents significant opportunities to enhance convenience, safety, and risk management for homeowners and service providers. However, secure onboarding-provisioning credentials and establishing trust with cloud platforms-remains a considerable challenge. Traditional onboarding methods often rely on centralized Public Key Infrastructure (PKI) models and manufacturer-controlled keys, which introduce security risks and limit the user's digital sovereignty. These limitations hinder the widespread deployment of scalable IoT solutions. This paper presents a novel onboarding framework that builds upon existing network-layer onboarding techniques and extends them to the application layer to address these challenges. By integrating consortium blockchain technology, we propose a decentralized onboarding mechanism that enhances transparency, security, and monitoring for smart home architectures. The architecture supports device registration, key revocation, access control management, and risk detection through event-driven alerts across dedicated blockchain channels and smart contracts. To evaluate the framework, we formally model the protocol using the Tamarin Prover under the Dolev-Yao adversary model. The analysis focuses on authentication, token integrity, key confidentiality, and resilience over public channels. A prototype implementation demonstrates the system's viability in smart home settings, with verification completing in 0.34 seconds, highlighting its scalability and suitability for constrained devices and diverse stakeholders. Additionally, performance evaluation shows that the blockchain-based approach effectively handles varying workloads, maintains high throughput and low latency, and supports near real-time IoT data processing.
SoK: Large Language Model-Generated Textual Phishing Campaigns End-to-End Analysis of Generation, Characteristics, and Detection
Phishing is a pervasive form of social engineering in which attackers impersonate trusted entities to steal information or induce harmful actions. Text-based phishing dominates for its low cost, scalability, and concealability, advantages recently amplified by large language models (LLMs) that enable ``Phishing-as-a-Service'' attacks at scale within minutes. Despite the growing research into LLM-facilitated phishing attacks, consolidated systematic research on the phishing attack life cycle remains scarce. In this work, we present the first systematization of knowledge (SoK) on LLM-generated phishing, offering an end-to-end analysis that spans generation techniques, attack features, and mitigation strategies. We introduce Generation-Characterization-Defense (GenCharDef), which systematizes the ways in which LLM-generated phishing differs from traditional phishing across methodologies, security perspectives, data dependencies, and evaluation practices. This framework highlights unique challenges of LLM-driven phishing, providing a coherent foundation for understanding the evolving threat landscape and guiding the design of more resilient defenses.
comment: 13 pages, 3 tables, 4 figures
Time Tells All: Deanonymization of Blockchain RPC Users with Zero Transaction Fee (Extended Version)
Remote Procedure Call (RPC) services have become a primary gateway for users to access public blockchains. While they offer significant convenience, RPC services also introduce critical privacy challenges that remain insufficiently examined. Existing deanonymization attacks either do not apply to blockchain RPC users or incur costs like transaction fees assuming an active network eavesdropper. In this paper, we propose a novel deanonymization attack that can link an IP address of a RPC user to this user's blockchain pseudonym. Our analysis reveals a temporal correlation between the timestamps of transaction confirmations recorded on the public ledger and those of TCP packets sent by the victim when querying transaction status. We assume a strong passive adversary with access to network infrastructure, capable of monitoring traffic at network border routers or Internet exchange points. By monitoring network traffic and analyzing public ledgers, the attacker can link the IP address of the TCP packet to the pseudonym of the transaction initiator by exploiting the temporal correlation. This deanonymization attack incurs zero transaction fee. We mathematically model and analyze the attack method, perform large-scale measurements of blockchain ledgers, and conduct real-world attacks to validate the attack. Our attack achieves a high success rate of over 95% against normal RPC users on various blockchain networks, including Ethereum, Bitcoin and Solana.
RepoMark: A Code Usage Auditing Framework for Code Large Language Models
The rapid development of Large Language Models (LLMs) for code generation has transformed software development by automating coding tasks with unprecedented efficiency. However, the training of these models on open-source code repositories (e.g., from GitHub) raises critical ethical and legal concerns, particularly regarding data authorization and open-source license compliance. Developers are increasingly questioning whether model trainers have obtained proper authorization before using repositories for training, especially given the lack of transparency in data collection. To address these concerns, we propose a novel data marking framework RepoMark to audit the data usage of code LLMs. Our method enables repository owners to verify whether their code has been used in training, while ensuring semantic preservation, imperceptibility, and theoretical false detection rate (FDR) guarantees. By generating multiple semantically equivalent code variants, RepoMark introduces data marks into the code files, and during detection, RepoMark leverages a novel ranking-based hypothesis test to detect memorization within the model. Compared to prior data auditing approaches, RepoMark significantly enhances sample efficiency, allowing effective auditing even when the user's repository possesses only a small number of code files. Experiments demonstrate that RepoMark achieves a detection success rate over 90\% on small code repositories under a strict FDR guarantee of 5\%. This represents a significant advancement over existing data marking techniques, all of which only achieve accuracy below 55\% under identical settings. This further validates RepoMark as a robust, theoretically sound, and promising solution for enhancing transparency in code LLM training, which can safeguard the rights of repository owners.
An Empirical Study of Vulnerable Package Dependencies in LLM Repositories
Large language models (LLMs) have developed rapidly in recent years, revolutionizing various fields. Despite their widespread success, LLMs heavily rely on external code dependencies from package management systems, creating a complex and interconnected LLM dependency supply chain. Vulnerabilities in dependencies can expose LLMs to security risks. While existing research predominantly focuses on model-level security threats, vulnerabilities within the LLM dependency supply chain have been overlooked. To fill this gap, we conducted an empirical analysis of 52 open-source LLMs, examining their third-party dependencies and associated vulnerabilities. We then explored activities within the LLM repositories to understand how maintainers manage third-party vulnerabilities in practice. Finally, we compared third-party dependency vulnerabilities in the LLM ecosystem to those in the Python ecosystem. Our results show that half of the vulnerabilities in the LLM ecosystem remain undisclosed for more than 56.2 months, significantly longer than those in the Python ecosystem. Additionally, 75.8% of LLMs include vulnerable dependencies in their configuration files. This study advances the understanding of LLM supply chain risks, provides insights for practitioners, and highlights potential directions for improving the security of the LLM supply chain.
zkLoRA: Fine-Tuning Large Language Models with Verifiable Security via Zero-Knowledge Proofs
Fine-tuning large language models (LLMs) is crucial for adapting them to specific tasks, yet it remains computationally demanding and raises concerns about correctness and privacy, particularly in untrusted environments. Although parameter-efficient methods like Low-Rank Adaptation (LoRA) significantly reduce resource requirements, ensuring the security and verifiability of fine-tuning under zero-knowledge constraints remains an unresolved challenge. To address this, we introduce zkLoRA, the first framework to integrate LoRA fine-tuning with zero-knowledge proofs (ZKPs), achieving provable security and correctness. zkLoRA employs advanced cryptographic techniques -- such as lookup arguments, sumcheck protocols, and polynomial commitments -- to verify both arithmetic and non-arithmetic operations in Transformer-based architectures. The framework provides end-to-end verifiability for forward propagation, backward propagation, and parameter updates during LoRA fine-tuning, while safeguarding the privacy of model parameters and training data. Leveraging GPU-based implementations, zkLoRA demonstrates practicality and efficiency through experimental validation on open-source LLMs like LLaMA, scaling up to 13 billion parameters. By combining parameter-efficient fine-tuning with ZKPs, zkLoRA bridges a critical gap, enabling secure and trustworthy deployment of LLMs in sensitive or untrusted environments.
Risks and Compliance with the EU's Core Cyber Security Legislation
The European Union (EU) has long favored a risk-based approach to regulation. Such an approach is also used in recent cyber security legislation enacted in the EU. Risks are also inherently related to compliance with the new legislation. Objective: The paper investigates how risks are framed in the EU's five core cyber security legislative acts, whether the framings indicate convergence or divergence between the acts and their risk concepts, and what qualifying words and terms are used when describing the legal notions of risks. Method : The paper's methodology is based on qualitative legal interpretation and taxonomy-building. Results: The five acts have an encompassing coverage of different cyber security risks, including but not limited to risks related to technical, organizational, and human security as well as those not originating from man-made actions. Both technical aspects and assets are used to frame the legal risk notions in many of the legislative acts. A threat-centric viewpoint is also present in one of the acts. Notable gaps are related to acceptable risks, non-probabilistic risks, and residual risks. Conclusion: The EU's new cyber security legislation has significantly extended the risk-based approach to regulations. At the same time, complexity and compliance burden have increased. With this point in mind, the paper concludes with a few practical takeaways about means to deal with compliance and research it.
comment: Submitted to IST (VSI:RegCompliance in SE)
LLM-driven Provenance Forensics for Threat Investigation and Detection
We introduce PROVSEEK, an LLM-powered agentic framework for automated provenance-driven forensic analysis and threat intelligence extraction. PROVSEEK employs specialized toolchains to dynamically retrieve relevant context by generating precise, context-aware queries that fuse a vectorized threat report knowledge base with data from system provenance databases. The framework resolves provenance queries, orchestrates multiple role-specific agents to mitigate hallucinations, and synthesizes structured, ground-truth verifiable forensic summaries. By combining agent orchestration with Retrieval-Augmented Generation (RAG) and chain-of-thought (CoT) reasoning, PROVSEEK enables adaptive multi-step analysis that iteratively refines hypotheses, verifies supporting evidence, and produces scalable, interpretable forensic explanations of attack behaviors. By combining provenance data with agentic reasoning, PROVSEEK establishes a new paradigm for grounded agentic forecics to investigate APTs. We conduct a comprehensive evaluation on publicly available DARPA datasets, demonstrating that PROVSEEK outperforms retrieval-based methods for intelligence extraction task, achieving a 34% improvement in contextual precision/recall; and for threat detection task, PROVSEEK achieves 22%/29% higher precision/recall compared to both a baseline agentic AI approach and State-Of-The-Art (SOTA) Provenance-based Intrusion Detection System (PIDS).
Locus: Agentic Predicate Synthesis for Directed Fuzzing
Directed fuzzing aims to find program inputs that lead to specified target program states. It has broad applications, such as debugging system crashes, confirming reported bugs, and generating exploits for potential vulnerabilities. This task is inherently challenging because target states are often deeply nested in the program, while the search space manifested by numerous possible program inputs is prohibitively large. Existing approaches rely on branch distances or manually-specified constraints to guide the search; however, the branches alone are often insufficient to precisely characterize progress toward reaching the target states, while the manually specified constraints are often tailored for specific bug types and thus difficult to generalize to diverse target states and programs. We present Locus, a novel framework to improve the efficiency of directed fuzzing. Our key insight is to synthesize predicates to capture fuzzing progress as semantically meaningful intermediate states, serving as milestones towards reaching the target states. When used to instrument the program under fuzzing, they can reject executions unlikely to reach the target states, while providing additional coverage guidance. To automate this task and generalize to diverse programs, Locus features an agentic framework with program analysis tools to synthesize and iteratively refine the candidate predicates, while ensuring the predicates strictly relax the target states to prevent false rejections via symbolic execution. Our evaluation shows that Locus substantially improves the efficiency of eight state-of-the-art fuzzers in discovering real-world vulnerabilities, achieving an average speedup of 41.6x. So far, Locus has found eight previously unpatched bugs, with one already acknowledged with a draft patch.
SAGA: A Security Architecture for Governing AI Agentic Systems
Large Language Model (LLM)-based agents increasingly interact, collaborate, and delegate tasks to one another autonomously with minimal human interaction. Industry guidelines for agentic system governance emphasize the need for users to maintain comprehensive control over their agents, mitigating potential damage from malicious agents. Several proposed agentic system designs address agent identity, authorization, and delegation, but remain purely theoretical, without concrete implementation and evaluation. Most importantly, they do not provide user-controlled agent management. To address this gap, we propose SAGA, a scalable Security Architecture for Governing Agentic systems, that offers user oversight over their agents' lifecycle. In our design, users register their agents with a central entity, the Provider, that maintains agent contact information, user-defined access control policies, and helps agents enforce these policies on inter-agent communication. We introduce a cryptographic mechanism for deriving access control tokens, that offers fine-grained control over an agent's interaction with other agents, providing formal security guarantees. We evaluate SAGA on several agentic tasks, using agents in different geolocations, and multiple on-device and cloud LLMs, demonstrating minimal performance overhead with no impact on underlying task utility in a wide range of conditions. Our architecture enables secure and trustworthy deployment of autonomous agents, accelerating the responsible adoption of this technology in sensitive environments.
Asynchronous Approximate Agreement with Quadratic Communication SC 2025
We consider an asynchronous network of $n$ message-sending parties, up to $t$ of which are byzantine. We study approximate agreement, where the parties obtain approximately equal outputs in the convex hull of their inputs. In their seminal work, Abraham, Amit and Dolev [OPODIS '04] solve this problem in $\mathbb{R}$ with the optimal resilience $t < \frac{n}{3}$ with a protocol where each party reliably broadcasts a value in every iteration. This takes $\Theta(n^2)$ messages per reliable broadcast, or $\Theta(n^3)$ messages per iteration. In this work, we forgo reliable broadcast to achieve asynchronous approximate agreement against $t < \frac{n}{3}$ faults with a quadratic communication. In a tree with the maximum degree $\Delta$ and the centroid decomposition height $h$, we achieve edge agreement in at most $6h + 1$ rounds with $\mathcal{O}(n^2)$ messages of size $\mathcal{O}(\log \Delta + \log h)$ per round. We do this by designing a 6-round multivalued 2-graded consensus protocol and using it to recursively reduce the task to edge agreement in a subtree with a smaller centroid decomposition height. Then, we achieve edge agreement in the infinite path $\mathbb{Z}$, again with the help of 2-graded consensus. Finally, we show that our edge agreement protocol enables $\varepsilon$-agreement in $\mathbb{R}$ in $6\log_2\frac{M}{\varepsilon} + \mathcal{O}(\log \log \frac{M}{\varepsilon})$ rounds with $\mathcal{O}(n^2 \log \frac{M}{\varepsilon})$ messages and $\mathcal{O}(n^2\log \frac{M}{\varepsilon}\log \log \frac{M}{\varepsilon})$ bits of communication, where $M$ is the maximum non-byzantine input magnitude.
comment: 26 pages, full version of a DISC 2025 brief announcement
Publish to Perish: Prompt Injection Attacks on LLM-Assisted Peer Review
Large Language Models (LLMs) are increasingly being integrated into the scientific peer-review process, raising new questions about their reliability and resilience to manipulation. In this work, we investigate the potential for hidden prompt injection attacks, where authors embed adversarial text within a paper's PDF to influence the LLM-generated review. We begin by formalising three distinct threat models that envision attackers with different motivations -- not all of which implying malicious intent. For each threat model, we design adversarial prompts that remain invisible to human readers yet can steer an LLM's output toward the author's desired outcome. Using a user study with domain scholars, we derive four representative reviewing prompts used to elicit peer reviews from LLMs. We then evaluate the robustness of our adversarial prompts across (i) different reviewing prompts, (ii) different commercial LLM-based systems, and (iii) different peer-reviewed papers. Our results show that adversarial prompts can reliably mislead the LLM, sometimes in ways that adversely affect a "honest-but-lazy" reviewer. Finally, we propose and empirically assess methods to reduce detectability of adversarial prompts under automated content checks.
CLUE-MARK: Watermarking Diffusion Models using CLWE
As AI-generated images become widespread, reliable watermarking is essential for content verification, copyright enforcement, and combating disinformation. Existing techniques rely on heuristic approaches and lack formal guarantees of undetectability, making them vulnerable to steganographic attacks that can expose or erase the watermark. Additionally, these techniques often degrade output quality by introducing perceptible changes, which is not only undesirable but an important barrier to adoption in practice. In this work, we introduce CLUE-Mark, the first provably undetectable watermarking scheme for diffusion models. CLUE-Mark requires no changes to the model being watermarked, is computationally efficient, and because it is provably undetectable is guaranteed to have no impact on model output quality. Our approach leverages the Continuous Learning With Errors (CLWE) problem -- a cryptographically hard lattice problem -- to embed watermarks in the latent noise vectors used by diffusion models. By proving undetectability via reduction from a cryptographically hard problem we ensure not only that the watermark is imperceptible to human observers or adhoc heuristics, but to \emph{any} efficient detector that does not have the secret key. CLUE-Mark allows multiple keys to be embedded, enabling traceability of images to specific users without altering model parameters. Empirical evaluations on state-of-the-art diffusion models confirm that CLUE-Mark achieves high recoverability, preserves image quality, and is robust to minor perturbations such JPEG compression and brightness adjustments. Uniquely, CLUE-Mark cannot be detected nor removed by recent steganographic attacks.
RevPRAG: Revealing Poisoning Attacks in Retrieval-Augmented Generation through LLM Activation Analysis EMNLP 2025
Retrieval-Augmented Generation (RAG) enriches the input to LLMs by retrieving information from the relevant knowledge database, enabling them to produce responses that are more accurate and contextually appropriate. It is worth noting that the knowledge database, being sourced from publicly available channels such as Wikipedia, inevitably introduces a new attack surface. RAG poisoning involves injecting malicious texts into the knowledge database, ultimately leading to the generation of the attacker's target response (also called poisoned response). However, there are currently limited methods available for detecting such poisoning attacks. We aim to bridge the gap in this work. Particularly, we introduce RevPRAG, a flexible and automated detection pipeline that leverages the activations of LLMs for poisoned response detection. Our investigation uncovers distinct patterns in LLMs' activations when generating correct responses versus poisoned responses. Our results on multiple benchmark datasets and RAG architectures show our approach could achieve 98% true positive rate, while maintaining false positive rates close to 1%.
comment: Accepted to Findings of EMNLP 2025
On the Adversarial Robustness of Spiking Neural Networks Trained by Local Learning
Recent research has shown the vulnerability of Spiking Neural Networks (SNNs) under adversarial examples that are nearly indistinguishable from clean data in the context of frame-based and event-based information. The majority of these studies are constrained in generating adversarial examples using Backpropagation Through Time (BPTT), a gradient-based method which lacks biological plausibility. In contrast, local learning methods, which relax many of BPTT's constraints, remain under-explored in the context of adversarial attacks. To address this problem, we examine adversarial robustness in SNNs through the framework of four types of training algorithms. We provide an in-depth analysis of the ineffectiveness of gradient-based adversarial attacks to generate adversarial instances in this scenario. To overcome these limitations, we introduce a hybrid adversarial attack paradigm that leverages the transferability of adversarial instances. The proposed hybrid approach demonstrates superior performance, outperforming existing adversarial attack methods. Furthermore, the generalizability of the method is assessed under multi-step adversarial attacks, adversarial attacks in black-box FGSM scenarios, and within the non-spiking domain.
Survey of Privacy Threats and Countermeasures in Federated Learning
Federated learning is widely considered to be as a privacy-aware learning method because no training data is exchanged directly between clients. Nevertheless, there are threats to privacy in federated learning, and privacy countermeasures have been studied. However, we note that common and unique privacy threats among typical types of federated learning have not been categorized and described in a comprehensive and specific way. In this paper, we describe privacy threats and countermeasures for the typical types of federated learning; horizontal federated learning, vertical federated learning, and transfer federated learning.
comment: The revised paper has been accepted as a full paper for presentation at The 3rd IEEE International Conference on Federated Learning Technologies and Applications (FLTA25)
Artificial Intelligence 130
The Demon is in Ambiguity: Revisiting Situation Recognition with Single Positive Multi-Label Learning ICDM 2025
Context recognition (SR) is a fundamental task in computer vision that aims to extract structured semantic summaries from images by identifying key events and their associated entities. Specifically, given an input image, the model must first classify the main visual events (verb classification), then identify the participating entities and their semantic roles (semantic role labeling), and finally localize these entities in the image (semantic role localization). Existing methods treat verb classification as a single-label problem, but we show through a comprehensive analysis that this formulation fails to address the inherent ambiguity in visual event recognition, as multiple verb categories may reasonably describe the same image. This paper makes three key contributions: First, we reveal through empirical analysis that verb classification is inherently a multi-label problem due to the ubiquitous semantic overlap between verb categories. Second, given the impracticality of fully annotating large-scale datasets with multiple labels, we propose to reformulate verb classification as a single positive multi-label learning (SPMLL) problem - a novel perspective in SR research. Third, we design a comprehensive multi-label evaluation benchmark for SR that is carefully designed to fairly evaluate model performance in a multi-label setting. To address the challenges of SPMLL, we futher develop the Graph Enhanced Verb Multilayer Perceptron (GE-VerbMLP), which combines graph neural networks to capture label correlations and adversarial training to optimize decision boundaries. Extensive experiments on real-world datasets show that our approach achieves more than 3\% MAP improvement while remaining competitive on traditional top-1 and top-5 accuracy metrics.
comment: Accepted by ICDM 2025
Automated Clinical Problem Detection from SOAP Notes using a Collaborative Multi-Agent LLM Architecture
Accurate interpretation of clinical narratives is critical for patient care, but the complexity of these notes makes automation challenging. While Large Language Models (LLMs) show promise, single-model approaches can lack the robustness required for high-stakes clinical tasks. We introduce a collaborative multi-agent system (MAS) that models a clinical consultation team to address this gap. The system is tasked with identifying clinical problems by analyzing only the Subjective (S) and Objective (O) sections of SOAP notes, simulating the diagnostic reasoning process of synthesizing raw data into an assessment. A Manager agent orchestrates a dynamically assigned team of specialist agents who engage in a hierarchical, iterative debate to reach a consensus. We evaluated our MAS against a single-agent baseline on a curated dataset of 420 MIMIC-III notes. The dynamic multi-agent configuration demonstrated consistently improved performance in identifying congestive heart failure, acute kidney injury, and sepsis. Qualitative analysis of the agent debates reveals that this structure effectively surfaces and weighs conflicting evidence, though it can occasionally be susceptible to groupthink. By modeling a clinical team's reasoning process, our system offers a promising path toward more accurate, robust, and interpretable clinical decision support tools.
comment: Accepted to The 16th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (ACM-BCB 2025)(Poster Paper)
Tree-Guided Diffusion Planner
Planning with pretrained diffusion models has emerged as a promising approach for solving test-time guided control problems. However, standard gradient guidance typically performs optimally under convex and differentiable reward landscapes, showing substantially reduced effectiveness in real-world scenarios involving non-convex objectives, non-differentiable constraints, and multi-reward structures. Furthermore, recent supervised planning approaches require task-specific training or value estimators, which limits test-time flexibility and zero-shot generalization. We propose a Tree-guided Diffusion Planner (TDP), a zero-shot test-time planning framework that balances exploration and exploitation through structured trajectory generation. We frame test-time planning as a tree search problem using a bi-level sampling process: (1) diverse parent trajectories are produced via training-free particle guidance to encourage broad exploration, and (2) sub-trajectories are refined through fast conditional denoising guided by task objectives. TDP addresses the limitations of gradient guidance by exploring diverse trajectory regions and harnessing gradient information across this expanded solution space using only pretrained models and test-time reward signals. We evaluate TDP on three diverse tasks: maze gold-picking, robot arm block manipulation, and AntMaze multi-goal exploration. TDP consistently outperforms state-of-the-art approaches on all tasks. The project page can be found at: tree-diffusion-planner.github.io.
comment: 20 pages, 11 figures, 14 tables (main paper + appendix) / under review / project page will be available after the paper becomes public in arxiv
DynaMark: A Reinforcement Learning Framework for Dynamic Watermarking in Industrial Machine Tool Controllers
Industry 4.0's highly networked Machine Tool Controllers (MTCs) are prime targets for replay attacks that use outdated sensor data to manipulate actuators. Dynamic watermarking can reveal such tampering, but current schemes assume linear-Gaussian dynamics and use constant watermark statistics, making them vulnerable to the time-varying, partly proprietary behavior of MTCs. We close this gap with DynaMark, a reinforcement learning framework that models dynamic watermarking as a Markov decision process (MDP). It learns an adaptive policy online that dynamically adapts the covariance of a zero-mean Gaussian watermark using available measurements and detector feedback, without needing system knowledge. DynaMark maximizes a unique reward function balancing control performance, energy consumption, and detection confidence dynamically. We develop a Bayesian belief updating mechanism for real-time detection confidence in linear systems. This approach, independent of specific system assumptions, underpins the MDP for systems with linear dynamics. On a Siemens Sinumerik 828D controller digital twin, DynaMark achieves a reduction in watermark energy by 70% while preserving the nominal trajectory, compared to constant variance baselines. It also maintains an average detection delay equivalent to one sampling interval. A physical stepper-motor testbed validates these findings, rapidly triggering alarms with less control performance decline and exceeding existing benchmarks.
TMUAD: Enhancing Logical Capabilities in Unified Anomaly Detection Models with a Text Memory Bank
Anomaly detection, which aims to identify anomalies deviating from normal patterns, is challenging due to the limited amount of normal data available. Unlike most existing unified methods that rely on carefully designed image feature extractors and memory banks to capture logical relationships between objects, we introduce a text memory bank to enhance the detection of logical anomalies. Specifically, we propose a Three-Memory framework for Unified structural and logical Anomaly Detection (TMUAD). First, we build a class-level text memory bank for logical anomaly detection by the proposed logic-aware text extractor, which can capture rich logical descriptions of objects from input images. Second, we construct an object-level image memory bank that preserves complete object contours by extracting features from segmented objects. Third, we employ visual encoders to extract patch-level image features for constructing a patch-level memory bank for structural anomaly detection. These three complementary memory banks are used to retrieve and compare normal images that are most similar to the query image, compute anomaly scores at multiple levels, and fuse them into a final anomaly score. By unifying structural and logical anomaly detection through collaborative memory banks, TMUAD achieves state-of-the-art performance across seven publicly available datasets involving industrial and medical domains. The model and code are available at https://github.com/SIA-IDE/TMUAD.
MoE-Health: A Mixture of Experts Framework for Robust Multimodal Healthcare Prediction
Healthcare systems generate diverse multimodal data, including Electronic Health Records (EHR), clinical notes, and medical images. Effectively leveraging this data for clinical prediction is challenging, particularly as real-world samples often present with varied or incomplete modalities. Existing approaches typically require complete modality data or rely on manual selection strategies, limiting their applicability in real-world clinical settings where data availability varies across patients and institutions. To address these limitations, we propose MoE-Health, a novel Mixture of Experts framework designed for robust multimodal fusion in healthcare prediction. MoE-Health architecture is specifically developed to handle samples with differing modalities and improve performance on critical clinical tasks. By leveraging specialized expert networks and a dynamic gating mechanism, our approach dynamically selects and combines relevant experts based on available data modalities, enabling flexible adaptation to varying data availability scenarios. We evaluate MoE-Health on the MIMIC-IV dataset across three critical clinical prediction tasks: in-hospital mortality prediction, long length of stay, and hospital readmission prediction. Experimental results demonstrate that MoE-Health achieves superior performance compared to existing multimodal fusion methods while maintaining robustness across different modality availability patterns. The framework effectively integrates multimodal information, offering improved predictive performance and robustness in handling heterogeneous and incomplete healthcare data, making it particularly suitable for deployment in diverse healthcare environments with heterogeneous data availability.
comment: Accepted to The 16th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics (ACM-BCB 2025)
Going over Fine Web with a Fine-Tooth Comb: Technical Report of Indexing Fine Web for Problematic Content Search and Retrieval
Large language models (LLMs) rely heavily on web-scale datasets like Common Crawl, which provides over 80\% of training data for some modern models. However, the indiscriminate nature of web crawling raises challenges in data quality, safety, and ethics. Despite the critical importance of training data quality, prior research on harmful content has been limited to small samples due to computational constraints. This project presents a framework for indexing and analyzing LLM training datasets using an ElasticSearch-based pipeline. We apply it to SwissAI's FineWeb-2 corpus (1.5TB, four languages), achieving fast query performance--most searches in milliseconds, all under 2 seconds. Our work demonstrates real-time dataset analysis, offering practical tools for safer, more accountable AI systems.
PiCSAR: Probabilistic Confidence Selection And Ranking
Best-of-n sampling improves the accuracy of large language models (LLMs) and large reasoning models (LRMs) by generating multiple candidate solutions and selecting the one with the highest reward. The key challenge for reasoning tasks is designing a scoring function that can identify correct reasoning chains without access to ground-truth answers. We propose Probabilistic Confidence Selection And Ranking (PiCSAR): a simple, training-free method that scores each candidate generation using the joint log-likelihood of the reasoning and final answer. The joint log-likelihood of the reasoning and final answer naturally decomposes into reasoning confidence and answer confidence. PiCSAR achieves substantial gains across diverse benchmarks (+10.18 on MATH500, +9.81 on AIME2025), outperforming baselines with at least 2x fewer samples in 16 out of 20 comparisons. Our analysis reveals that correct reasoning chains exhibit significantly higher reasoning and answer confidence, justifying the effectiveness of PiCSAR.
Benchmarking GPT-5 in Radiation Oncology: Measurable Gains, but Persistent Need for Expert Oversight
Introduction: Large language models (LLM) have shown great potential in clinical decision support. GPT-5 is a novel LLM system that has been specifically marketed towards oncology use. Methods: Performance was assessed using two complementary benchmarks: (i) the ACR Radiation Oncology In-Training Examination (TXIT, 2021), comprising 300 multiple-choice items, and (ii) a curated set of 60 authentic radiation oncologic vignettes representing diverse disease sites and treatment indications. For the vignette evaluation, GPT-5 was instructed to generate concise therapeutic plans. Four board-certified radiation oncologists rated correctness, comprehensiveness, and hallucinations. Inter-rater reliability was quantified using Fleiss' \k{appa}. Results: On the TXIT benchmark, GPT-5 achieved a mean accuracy of 92.8%, outperforming GPT-4 (78.8%) and GPT-3.5 (62.1%). Domain-specific gains were most pronounced in Dose and Diagnosis. In the vignette evaluation, GPT-5's treatment recommendations were rated highly for correctness (mean 3.24/4, 95% CI: 3.11-3.38) and comprehensiveness (3.59/4, 95% CI: 3.49-3.69). Hallucinations were rare with no case reaching majority consensus for their presence. Inter-rater agreement was low (Fleiss' \k{appa} 0.083 for correctness), reflecting inherent variability in clinical judgment. Errors clustered in complex scenarios requiring precise trial knowledge or detailed clinical adaptation. Discussion: GPT-5 clearly outperformed prior model variants on the radiation oncology multiple-choice benchmark. Although GPT-5 exhibited favorable performance in generating real-world radiation oncology treatment recommendations, correctness ratings indicate room for further improvement. While hallucinations were infrequent, the presence of substantive errors underscores that GPT-5-generated recommendations require rigorous expert oversight before clinical implementation.
comment: Under review in Frontiers in Artificial Intelligence
Unsupervised Video Continual Learning via Non-Parametric Deep Embedded Clustering BMVC 2025
We propose a realistic scenario for the unsupervised video learning where neither task boundaries nor labels are provided when learning a succession of tasks. We also provide a non-parametric learning solution for the under-explored problem of unsupervised video continual learning. Videos represent a complex and rich spatio-temporal media information, widely used in many applications, but which have not been sufficiently explored in unsupervised continual learning. Prior studies have only focused on supervised continual learning, relying on the knowledge of labels and task boundaries, while having labeled data is costly and not practical. To address this gap, we study the unsupervised video continual learning (uVCL). uVCL raises more challenges due to the additional computational and memory requirements of processing videos when compared to images. We introduce a general benchmark experimental protocol for uVCL by considering the learning of unstructured video data categories during each task. We propose to use the Kernel Density Estimation (KDE) of deep embedded video features extracted by unsupervised video transformer networks as a non-parametric probabilistic representation of the data. We introduce a novelty detection criterion for the incoming new task data, dynamically enabling the expansion of memory clusters, aiming to capture new knowledge when learning a succession of tasks. We leverage the use of transfer learning from the previous tasks as an initial state for the knowledge transfer to the current learning task. We found that the proposed methodology substantially enhances the performance of the model when successively learning many tasks. We perform in-depth evaluations on three standard video action recognition datasets, including UCF101, HMDB51, and Something-to-Something V2, without using any labels or class boundaries.
comment: Accepted to The 36th British Machine Vision Conference (BMVC 2025), Sheffield, UK
Reasoning-Intensive Regression
AI researchers and practitioners increasingly apply large language models (LLMs) to what we call reasoning-intensive regression (RiR), i.e. deducing subtle numerical properties from text. Unlike standard language regression tasks, e.g. for sentiment or similarity, RiR often appears instead in ad-hoc problems like rubric-based scoring or domain-specific retrieval, where much deeper analysis of text is required while only limited task-specific training data and computation are available. We cast three realistic problems as RiR tasks to establish an initial benchmark, and use that to test our hypothesis that prompting frozen LLMs and finetuning Transformer encoders via gradient descent will both often struggle in RiR. We then propose MENTAT, a simple and lightweight method that combines batch-reflective prompt optimization with neural ensemble learning. MENTAT achieves up to 65% improvement over both baselines, though substantial room remains for future advances in RiR.
Orientability of Causal Relations in Time Series using Summary Causal Graphs and Faithful Distributions
Understanding causal relations between temporal variables is a central challenge in time series analysis, particularly when the full causal structure is unknown. Even when the full causal structure cannot be fully specified, experts often succeed in providing a high-level abstraction of the causal graph, known as a summary causal graph, which captures the main causal relations between different time series while abstracting away micro-level details. In this work, we present conditions that guarantee the orientability of micro-level edges between temporal variables given the background knowledge encoded in a summary causal graph and assuming having access to a faithful and causally sufficient distribution with respect to the true unknown graph. Our results provide theoretical guarantees for edge orientation at the micro-level, even in the presence of cycles or bidirected edges at the macro-level. These findings offer practical guidance for leveraging SCGs to inform causal discovery in complex temporal systems and highlight the value of incorporating expert knowledge to improve causal inference from observational time series data.
Neural Network Acceleration on MPSoC board: Integrating SLAC's SNL, Rogue Software and Auto-SNL
The LCLS-II Free Electron Laser (FEL) will generate X-ray pulses for beamline experiments at rates of up to 1~MHz, with detectors producing data throughputs exceeding 1 TB/s. Managing such massive data streams presents significant challenges, as transmission and storage infrastructures become prohibitively expensive. Machine learning (ML) offers a promising solution for real-time data reduction, but conventional implementations introduce excessive latency, making them unsuitable for high-speed experimental environments. To address these challenges, SLAC developed the SLAC Neural Network Library (SNL), a specialized framework designed to deploy real-time ML inference models on Field-Programmable Gate Arrays (FPGA). SNL's key feature is the ability to dynamically update model weights without requiring FPGA resynthesis, enhancing flexibility for adaptive learning applications. To further enhance usability and accessibility, we introduce Auto-SNL, a Python extension that streamlines the process of converting Python-based neural network models into SNL-compatible high-level synthesis code. This paper presents a benchmark comparison against hls4ml, the current state-of-the-art tool, across multiple neural network architectures, fixed-point precisions, and synthesis configurations targeting a Xilinx ZCU102 FPGA. The results showed that SNL achieves competitive or superior latency in most tested architectures, while in some cases also offering FPGA resource savings. This adaptation demonstrates SNL's versatility, opening new opportunities for researchers and academics in fields such as high-energy physics, medical imaging, robotics, and many more.
Developer Insights into Designing AI-Based Computer Perception Tools
Artificial intelligence (AI)-based computer perception (CP) technologies use mobile sensors to collect behavioral and physiological data for clinical decision-making. These tools can reshape how clinical knowledge is generated and interpreted. However, effective integration of these tools into clinical workflows depends on how developers balance clinical utility with user acceptability and trustworthiness. Our study presents findings from 20 in-depth interviews with developers of AI-based CP tools. Interviews were transcribed and inductive, thematic analysis was performed to identify 4 key design priorities: 1) to account for context and ensure explainability for both patients and clinicians; 2) align tools with existing clinical workflows; 3) appropriately customize to relevant stakeholders for usability and acceptability; and 4) push the boundaries of innovation while aligning with established paradigms. Our findings highlight that developers view themselves as not merely technical architects but also ethical stewards, designing tools that are both acceptable by users and epistemically responsible (prioritizing objectivity and pushing clinical knowledge forward). We offer the following suggestions to help achieve this balance: documenting how design choices around customization are made, defining limits for customization choices, transparently conveying information about outputs, and investing in user training. Achieving these goals will require interdisciplinary collaboration between developers, clinicians, and ethicists.
comment: 15 pages
CAD2DMD-SET: Synthetic Generation Tool of Digital Measurement Device CAD Model Datasets for fine-tuning Large Vision-Language Models
Recent advancements in Large Vision-Language Models (LVLMs) have demonstrated impressive capabilities across various multimodal tasks. They continue, however, to struggle with trivial scenarios such as reading values from Digital Measurement Devices (DMDs), particularly in real-world conditions involving clutter, occlusions, extreme viewpoints, and motion blur; common in head-mounted cameras and Augmented Reality (AR) applications. Motivated by these limitations, this work introduces CAD2DMD-SET, a synthetic data generation tool designed to support visual question answering (VQA) tasks involving DMDs. By leveraging 3D CAD models, advanced rendering, and high-fidelity image composition, our tool produces diverse, VQA-labelled synthetic DMD datasets suitable for fine-tuning LVLMs. Additionally, we present DMDBench, a curated validation set of 1,000 annotated real-world images designed to evaluate model performance under practical constraints. Benchmarking three state-of-the-art LVLMs using Average Normalised Levenshtein Similarity (ANLS) and further fine-tuning LoRA's of these models with CAD2DMD-SET's generated dataset yielded substantial improvements, with InternVL showcasing a score increase of 200% without degrading on other tasks. This demonstrates that the CAD2DMD-SET training dataset substantially improves the robustness and performance of LVLMs when operating under the previously stated challenging conditions. The CAD2DMD-SET tool is expected to be released as open-source once the final version of this manuscript is prepared, allowing the community to add different measurement devices and generate their own datasets.
Freeze and Conquer: Reusable Ansatz for Solving the Traveling Salesman Problem
In this paper we present a variational algorithm for the Traveling Salesman Problem (TSP) that combines (i) a compact encoding of permutations, which reduces the qubit requirement too, (ii) an optimize-freeze-reuse strategy: where the circuit topology (``Ansatz'') is first optimized on a training instance by Simulated Annealing (SA), then ``frozen'' and re-used on novel instances, limited to a rapid re-optimization of only the circuit parameters. This pipeline eliminates costly structural research in testing, making the procedure immediately implementable on NISQ hardware. On a set of $40$ randomly generated symmetric instances that span $4 - 7$ cities, the resulting Ansatz achieves an average optimal trip sampling probability of $100\%$ for 4 city cases, $90\%$ for 5 city cases and $80\%$ for 6 city cases. With 7 cities the success rate drops markedly to an average of $\sim 20\%$, revealing the onset of scalability limitations of the proposed method. The results show robust generalization ability for moderate problem sizes and indicate how freezing the Ansatz can dramatically reduce time-to-solution without degrading solution quality. The paper also discusses scalability limitations, the impact of ``warm-start'' initialization of parameters, and prospects for extension to more complex problems, such as Vehicle Routing and Job-Shop Scheduling.
OptMark: Robust Multi-bit Diffusion Watermarking via Inference Time Optimization
Watermarking diffusion-generated images is crucial for copyright protection and user tracking. However, current diffusion watermarking methods face significant limitations: zero-bit watermarking systems lack the capacity for large-scale user tracking, while multi-bit methods are highly sensitive to certain image transformations or generative attacks, resulting in a lack of comprehensive robustness. In this paper, we propose OptMark, an optimization-based approach that embeds a robust multi-bit watermark into the intermediate latents of the diffusion denoising process. OptMark strategically inserts a structural watermark early to resist generative attacks and a detail watermark late to withstand image transformations, with tailored regularization terms to preserve image quality and ensure imperceptibility. To address the challenge of memory consumption growing linearly with the number of denoising steps during optimization, OptMark incorporates adjoint gradient methods, reducing memory usage from O(N) to O(1). Experimental results demonstrate that OptMark achieves invisible multi-bit watermarking while ensuring robust resilience against valuemetric transformations, geometric transformations, editing, and regeneration attacks.
PosterForest: Hierarchical Multi-Agent Collaboration for Scientific Poster Generation
We present a novel training-free framework, \textit{PosterForest}, for automated scientific poster generation. Unlike prior approaches, which largely neglect the hierarchical structure of scientific documents and the semantic integration of textual and visual elements, our method addresses both challenges directly. We introduce the \textit{Poster Tree}, a hierarchical intermediate representation that jointly encodes document structure and visual-textual relationships at multiple levels. Our framework employs a multi-agent collaboration strategy, where agents specializing in content summarization and layout planning iteratively coordinate and provide mutual feedback. This approach enables the joint optimization of logical consistency, content fidelity, and visual coherence. Extensive experiments on multiple academic domains show that our method outperforms existing baselines in both qualitative and quantitative evaluations. The resulting posters achieve quality closest to expert-designed ground truth and deliver superior information preservation, structural clarity, and user preference.
Entropy-Based Non-Invasive Reliability Monitoring of Convolutional Neural Networks
Convolutional Neural Networks (CNNs) have become the foundation of modern computer vision, achieving unprecedented accuracy across diverse image recognition tasks. While these networks excel on in-distribution data, they remain vulnerable to adversarial perturbations imperceptible input modifications that cause misclassification with high confidence. However, existing detection methods either require expensive retraining, modify network architecture, or degrade performance on clean inputs. Here we show that adversarial perturbations create immediate, detectable entropy signatures in CNN activations that can be monitored without any model modification. Using parallel entropy monitoring on VGG-16, we demonstrate that adversarial inputs consistently shift activation entropy by 7% in early convolutional layers, enabling 90% detection accuracy with false positives and false negative rates below 20%. The complete separation between clean and adversarial entropy distributions reveals that CNNs inherently encode distribution shifts in their activation patterns. This work establishes that CNN reliability can be assessed through activation entropy alone, enabling practical deployment of self-diagnostic vision systems that detect adversarial inputs in real-time without compromising original model performance.
comment: 8 pages, 3 figures, 2 tables
Why Stop at Words? Unveiling the Bigger Picture through Line-Level OCR
Conventional optical character recognition (OCR) techniques segmented each character and then recognized. This made them prone to error in character segmentation, and devoid of context to exploit language models. Advances in sequence to sequence translation in last decade led to modern techniques first detecting words and then inputting one word at a time to a model to directly output full words as sequence of characters. This allowed better utilization of language models and bypass error-prone character segmentation step. We observe that the above transition in style has moved the bottleneck in accuracy to word segmentation. Hence, in this paper, we propose a natural and logical progression from word level OCR to line-level OCR. The proposal allows to bypass errors in word detection, and provides larger sentence context for better utilization of language models. We show that the proposed technique not only improves the accuracy but also efficiency of OCR. Despite our thorough literature survey, we did not find any public dataset to train and benchmark such shift from word to line-level OCR. Hence, we also contribute a meticulously curated dataset of 251 English page images with line-level annotations. Our experimentation revealed a notable end-to-end accuracy improvement of 5.4%, underscoring the potential benefits of transitioning towards line-level OCR, especially for document images. We also report a 4 times improvement in efficiency compared to word-based pipelines. With continuous improvements in large language models, our methodology also holds potential to exploit such advances. Project Website: https://nishitanand.github.io/line-level-ocr-website
comment: 11 pages. Project Website: https://nishitanand.github.io/line-level-ocr-website
Harnessing IoT and Generative AI for Weather-Adaptive Learning in Climate Resilience Education
This paper introduces the Future Atmospheric Conditions Training System (FACTS), a novel platform that advances climate resilience education through place-based, adaptive learning experiences. FACTS combines real-time atmospheric data collected by IoT sensors with curated resources from a Knowledge Base to dynamically generate localized learning challenges. Learner responses are analyzed by a Generative AI powered server, which delivers personalized feedback and adaptive support. Results from a user evaluation indicate that participants found the system both easy to use and effective for building knowledge related to climate resilience. These findings suggest that integrating IoT and Generative AI into atmospherically adaptive learning technologies holds significant promise for enhancing educational engagement and fostering climate awareness.
Leveraging Imperfection with MEDLEY A Multi-Model Approach Harnessing Bias in Medical AI
Bias in medical artificial intelligence is conventionally viewed as a defect requiring elimination. However, human reasoning inherently incorporates biases shaped by education, culture, and experience, suggesting their presence may be inevitable and potentially valuable. We propose MEDLEY (Medical Ensemble Diagnostic system with Leveraged diversitY), a conceptual framework that orchestrates multiple AI models while preserving their diverse outputs rather than collapsing them into a consensus. Unlike traditional approaches that suppress disagreement, MEDLEY documents model-specific biases as potential strengths and treats hallucinations as provisional hypotheses for clinician verification. A proof-of-concept demonstrator was developed using over 30 large language models, creating a minimum viable product that preserved both consensus and minority views in synthetic cases, making diagnostic uncertainty and latent biases transparent for clinical oversight. While not yet a validated clinical tool, the demonstration illustrates how structured diversity can enhance medical reasoning under clinician supervision. By reframing AI imperfection as a resource, MEDLEY offers a paradigm shift that opens new regulatory, ethical, and innovation pathways for developing trustworthy medical AI systems.
A-MHA*: Anytime Multi-Heuristic A*
Designing good heuristic functions for graph search requires adequate domain knowledge. It is often easy to design heuristics that perform well and correlate with the underlying true cost-to-go values in certain parts of the search space but these may not be admissible throughout the domain thereby affecting the optimality guarantees of the search. Bounded suboptimal search using several such partially good but inadmissible heuristics was developed in Multi-Heuristic A* (MHA*). Although MHA* leverages multiple inadmissible heuristics to potentially generate a faster suboptimal solution, the original version does not improve the solution over time. It is a one shot algorithm that requires careful setting of inflation factors to obtain a desired one time solution. In this work, we tackle this issue by extending MHA* to an anytime version that finds a feasible suboptimal solution quickly and continually improves it until time runs out. Our work is inspired from the Anytime Repairing A* (ARA*) algorithm. We prove that our precise adaptation of ARA* concepts in the MHA* framework preserves the original suboptimal and completeness guarantees and enhances MHA* to perform in an anytime fashion. Furthermore, we report the performance of A-MHA* in 3-D path planning domain and sliding tiles puzzle and compare against MHA* and other anytime algorithms.
QZhou-Embedding Technical Report
We present QZhou-Embedding, a general-purpose contextual text embedding model with exceptional text representation capabilities. Built upon the Qwen2.5-7B-Instruct foundation model, we designed a unified multi-task framework comprising specialized data transformation and training strategies. The data transformation scheme enables the incorporation of more diverse textual training datasets, while the task-specific training strategies enhance model learning efficiency. We developed a data synthesis pipeline leveraging LLM API, incorporating techniques such as paraphrasing, augmentation, and hard negative example generation to improve the semantic richness and sample difficulty of the training set. Additionally, we employ a two-stage training strategy, comprising initial retrieval-focused pretraining followed by full-task fine-tuning, enabling the embedding model to extend its capabilities based on robust retrieval performance. Our model achieves state-of-the-art results on the MTEB and CMTEB benchmarks, ranking first on both leaderboards (August 27 2025), and simultaneously achieves state-of-the-art performance on tasks including reranking, clustering, etc. Our findings demonstrate that higher-quality, more diverse data is crucial for advancing retrieval model performance, and that leveraging LLMs generative capabilities can further optimize data quality for embedding model breakthroughs. Our model weights are released on HuggingFace under Apache 2.0 license. For reproducibility, we provide evaluation code and instructions on GitHub.
Integrating Large Language Models with Network Optimization for Interactive and Explainable Supply Chain Planning: A Real-World Case Study
This paper presents an integrated framework that combines traditional network optimization models with large language models (LLMs) to deliver interactive, explainable, and role-aware decision support for supply chain planning. The proposed system bridges the gap between complex operations research outputs and business stakeholder understanding by generating natural language summaries, contextual visualizations, and tailored key performance indicators (KPIs). The core optimization model addresses tactical inventory redistribution across a network of distribution centers for multi-period and multi-item, using a mixed-integer formulation. The technical architecture incorporates AI agents, RESTful APIs, and a dynamic user interface to support real-time interaction, configuration updates, and simulation-based insights. A case study demonstrates how the system improves planning outcomes by preventing stockouts, reducing costs, and maintaining service levels. Future extensions include integrating private LLMs, transfer learning, reinforcement learning, and Bayesian neural networks to enhance explainability, adaptability, and real-time decision-making.
Physics-Informed Spectral Modeling for Hyperspectral Imaging
We present PhISM, a physics-informed deep learning architecture that learns without supervision to explicitly disentangle hyperspectral observations and model them with continuous basis functions. \mname outperforms prior methods on several classification and regression benchmarks, requires limited labeled data, and provides additional insights thanks to interpretable latent representation.
Scalable Solution Methods for Dec-POMDPs with Deterministic Dynamics
Many high-level multi-agent planning problems, including multi-robot navigation and path planning, can be effectively modeled using deterministic actions and observations. In this work, we focus on such domains and introduce the class of Deterministic Decentralized POMDPs (Det-Dec-POMDPs). This is a subclass of Dec-POMDPs characterized by deterministic transitions and observations conditioned on the state and joint actions. We then propose a practical solver called Iterative Deterministic POMDP Planning (IDPP). This method builds on the classic Joint Equilibrium Search for Policies framework and is specifically optimized to handle large-scale Det-Dec-POMDPs that current Dec-POMDP solvers are unable to address efficiently.
Middo: Model-Informed Dynamic Data Optimization for Enhanced LLM Fine-Tuning via Closed-Loop Learning EMNLP 2025
Supervised Fine-Tuning (SFT) Large Language Models (LLM) fundamentally rely on high-quality training data. While data selection and data synthesis are two common strategies to improve data quality, existing approaches often face limitations in static dataset curation that fail to adapt to evolving model capabilities. In this paper, we introduce Middo, a self-evolving Model-informed dynamic data optimization framework that uses model-aware data selection and context-preserving data refinement. Unlike conventional one-off filtering/synthesis methods, our framework establishes a closed-loop optimization system: (1) A self-referential diagnostic module proactively identifies suboptimal samples through tri-axial model signals - loss patterns (complexity), embedding cluster dynamics (diversity), and self-alignment scores (quality); (2) An adaptive optimization engine then transforms suboptimal samples into pedagogically valuable training points while preserving semantic integrity; (3) This optimization process continuously evolves with model capability through dynamic learning principles. Experiments on multiple benchmarks demonstrate that our \method consistently enhances the quality of seed data and boosts LLM's performance with improving accuracy by 7.15% on average while maintaining the original dataset scale. This work establishes a new paradigm for sustainable LLM training through dynamic human-AI co-evolution of data and models. Our datasets, models, and code are coming soon.
comment: Accepted by EMNLP 2025 (main)
A Survey on Current Trends and Recent Advances in Text Anonymization
The proliferation of textual data containing sensitive personal information across various domains requires robust anonymization techniques to protect privacy and comply with regulations, while preserving data usability for diverse and crucial downstream tasks. This survey provides a comprehensive overview of current trends and recent advances in text anonymization techniques. We begin by discussing foundational approaches, primarily centered on Named Entity Recognition, before examining the transformative impact of Large Language Models, detailing their dual role as sophisticated anonymizers and potent de-anonymization threats. The survey further explores domain-specific challenges and tailored solutions in critical sectors such as healthcare, law, finance, and education. We investigate advanced methodologies incorporating formal privacy models and risk-aware frameworks, and address the specialized subfield of authorship anonymization. Additionally, we review evaluation frameworks, comprehensive metrics, benchmarks, and practical toolkits for real-world deployment of anonymization solutions. This review consolidates current knowledge, identifies emerging trends and persistent challenges, including the evolving privacy-utility trade-off, the need to address quasi-identifiers, and the implications of LLM capabilities, and aims to guide future research directions for both academics and practitioners in this field.
comment: Accepted at IEEE DSAA 2025
NSPDI-SNN: An efficient lightweight SNN based on nonlinear synaptic pruning and dendritic integration
Spiking neural networks (SNNs) are artificial neural networks based on simulated biological neurons and have attracted much attention in recent artificial intelligence technology studies. The dendrites in biological neurons have efficient information processing ability and computational power; however, the neurons of SNNs rarely match the complex structure of the dendrites. Inspired by the nonlinear structure and highly sparse properties of neuronal dendrites, in this study, we propose an efficient, lightweight SNN method with nonlinear pruning and dendritic integration (NSPDI-SNN). In this method, we introduce nonlinear dendritic integration (NDI) to improve the representation of the spatiotemporal information of neurons. We implement heterogeneous state transition ratios of dendritic spines and construct a new and flexible nonlinear synaptic pruning (NSP) method to achieve the high sparsity of SNN. We conducted systematic experiments on three benchmark datasets (DVS128 Gesture, CIFAR10-DVS, and CIFAR10) and extended the evaluation to two complex tasks (speech recognition and reinforcement learning-based maze navigation task). Across all tasks, NSPDI-SNN consistently achieved high sparsity with minimal performance degradation. In particular, our method achieved the best experimental results on all three event stream datasets. Further analysis showed that NSPDI significantly improved the efficiency of synaptic information transfer as sparsity increased. In conclusion, our results indicate that the complex structure and nonlinear computation of neuronal dendrites provide a promising approach for developing efficient SNN methods.
comment: 13 pages, 8 figures, 5 tables; This manuscript has been submitted for possible pulication
Revisiting Landmarks: Learning from Previous Plans to Generalize over Problem Instances
We propose a new framework for discovering landmarks that automatically generalize across a domain. These generalized landmarks are learned from a set of solved instances and describe intermediate goals for planning problems where traditional landmark extraction algorithms fall short. Our generalized landmarks extend beyond the predicates of a domain by using state functions that are independent of the objects of a specific problem and apply to all similar objects, thus capturing repetition. Based on these functions, we construct a directed generalized landmark graph that defines the landmark progression, including loop possibilities for repetitive subplans. We show how to use this graph in a heuristic to solve new problem instances of the same domain. Our results show that the generalized landmark graphs learned from a few small instances are also effective for larger instances in the same domain. If a loop that indicates repetition is identified, we see a significant improvement in heuristic performance over the baseline. Generalized landmarks capture domain information that is interpretable and useful to an automated planner. This information can be discovered from a small set of plans for the same domain.
Limitations of Physics-Informed Neural Networks: a Study on Smart Grid Surrogation
Physics-Informed Neural Networks (PINNs) present a transformative approach for smart grid modeling by integrating physical laws directly into learning frameworks, addressing critical challenges of data scarcity and physical consistency in conventional data-driven methods. This paper evaluates PINNs' capabilities as surrogate models for smart grid dynamics, comparing their performance against XGBoost, Random Forest, and Linear Regression across three key experiments: interpolation, cross-validation, and episodic trajectory prediction. By training PINNs exclusively through physics-based loss functions (enforcing power balance, operational constraints, and grid stability) we demonstrate their superior generalization, outperforming data-driven models in error reduction. Notably, PINNs maintain comparatively lower MAE in dynamic grid operations, reliably capturing state transitions in both random and expert-driven control scenarios, while traditional models exhibit erratic performance. Despite slight degradation in extreme operational regimes, PINNs consistently enforce physical feasibility, proving vital for safety-critical applications. Our results contribute to establishing PINNs as a paradigm-shifting tool for smart grid surrogation, bridging data-driven flexibility with first-principles rigor. This work advances real-time grid control and scalable digital twins, emphasizing the necessity of physics-aware architectures in mission-critical energy systems.
comment: Presented in PowerTech2025
EZ-Sort: Efficient Pairwise Comparison via Zero-Shot CLIP-Based Pre-Ordering and Human-in-the-Loop Sorting CIKM 2025
Pairwise comparison is often favored over absolute rating or ordinal classification in subjective or difficult annotation tasks due to its improved reliability. However, exhaustive comparisons require a massive number of annotations (O(n^2)). Recent work has greatly reduced the annotation burden (O(n log n)) by actively sampling pairwise comparisons using a sorting algorithm. We further improve annotation efficiency by (1) roughly pre-ordering items using the Contrastive Language-Image Pre-training (CLIP) model hierarchically without training, and (2) replacing easy, obvious human comparisons with automated comparisons. The proposed EZ-Sort first produces a CLIP-based zero-shot pre-ordering, then initializes bucket-aware Elo scores, and finally runs an uncertainty-guided human-in-the-loop MergeSort. Validation was conducted using various datasets: face-age estimation (FGNET), historical image chronology (DHCI), and retinal image quality assessment (EyePACS). It showed that EZ-Sort reduced human annotation cost by 90.5% compared to exhaustive pairwise comparisons and by 19.8% compared to prior work (when n = 100), while improving or maintaining inter-rater reliability. These results demonstrate that combining CLIP-based priors with uncertainty-aware sampling yields an efficient and scalable solution for pairwise ranking.
comment: 5 pages, 2 figures, Accepted at CIKM 2025 (ACM International Conference on Information and Knowledge Management)
What Data is Really Necessary? A Feasibility Study of Inference Data Minimization for Recommender Systems CIKM '25
Data minimization is a legal principle requiring personal data processing to be limited to what is necessary for a specified purpose. Operationalizing this principle for recommender systems, which rely on extensive personal data, remains a significant challenge. This paper conducts a feasibility study on minimizing implicit feedback inference data for such systems. We propose a novel problem formulation, analyze various minimization techniques, and investigate key factors influencing their effectiveness. We demonstrate that substantial inference data reduction is technically feasible without significant performance loss. However, its practicality is critically determined by two factors: the technical setting (e.g., performance targets, choice of model) and user characteristics (e.g., history size, preference complexity). Thus, while we establish its technical feasibility, we conclude that data minimization remains practically challenging and its dependence on the technical and user context makes a universal standard for data `necessity' difficult to implement.
comment: Accepted for publication at the 34th ACM International Conference on Information and Knowledge Management (CIKM '25), November 10-14, 2025, Seoul, Republic of Korea
Complete Gaussian Splats from a Single Image with Denoising Diffusion Models
Gaussian splatting typically requires dense observations of the scene and can fail to reconstruct occluded and unobserved areas. We propose a latent diffusion model to reconstruct a complete 3D scene with Gaussian splats, including the occluded parts, from only a single image during inference. Completing the unobserved surfaces of a scene is challenging due to the ambiguity of the plausible surfaces. Conventional methods use a regression-based formulation to predict a single "mode" for occluded and out-of-frustum surfaces, leading to blurriness, implausibility, and failure to capture multiple possible explanations. Thus, they often address this problem partially, focusing either on objects isolated from the background, reconstructing only visible surfaces, or failing to extrapolate far from the input views. In contrast, we propose a generative formulation to learn a distribution of 3D representations of Gaussian splats conditioned on a single input image. To address the lack of ground-truth training data, we propose a Variational AutoReconstructor to learn a latent space only from 2D images in a self-supervised manner, over which a diffusion model is trained. Our method generates faithful reconstructions and diverse samples with the ability to complete the occluded surfaces for high-quality 360-degree renderings.
comment: Main paper: 11 pages; Supplementary materials: 7 pages
HealthProcessAI: A Technical Framework and Proof-of-Concept for LLM-Enhanced Healthcare Process Mining
Process mining has emerged as a powerful analytical technique for understanding complex healthcare workflows. However, its application faces significant barriers, including technical complexity, a lack of standardized approaches, and limited access to practical training resources. We introduce HealthProcessAI, a GenAI framework designed to simplify process mining applications in healthcare and epidemiology by providing a comprehensive wrapper around existing Python (PM4PY) and R (bupaR) libraries. To address unfamiliarity and improve accessibility, the framework integrates multiple Large Language Models (LLMs) for automated process map interpretation and report generation, helping translate technical analyses into outputs that diverse users can readily understand. We validated the framework using sepsis progression data as a proof-of-concept example and compared the outputs of five state-of-the-art LLM models through the OpenRouter platform. To test its functionality, the framework successfully processed sepsis data across four proof-of-concept scenarios, demonstrating robust technical performance and its capability to generate reports through automated LLM analysis. LLM evaluation using five independent LLMs as automated evaluators revealed distinct model strengths: Claude Sonnet-4 and Gemini 2.5-Pro achieved the highest consistency scores (3.79/4.0 and 3.65/4.0) when evaluated by automated LLM assessors. By integrating multiple Large Language Models (LLMs) for automated interpretation and report generation, the framework addresses widespread unfamiliarity with process mining outputs, making them more accessible to clinicians, data scientists, and researchers. This structured analytics and AI-driven interpretation combination represents a novel methodological advance in translating complex process mining results into potentially actionable insights for healthcare applications.
Counterfactual Scenarios for Automated Planning KR 2025
Counterfactual Explanations (CEs) are a powerful technique used to explain Machine Learning models by showing how the input to a model should be minimally changed for the model to produce a different output. Similar proposals have been made in the context of Automated Planning, where CEs have been characterised in terms of minimal modifications to an existing plan that would result in the satisfaction of a different goal. While such explanations may help diagnose faults and reason about the characteristics of a plan, they fail to capture higher-level properties of the problem being solved. To address this limitation, we propose a novel explanation paradigm that is based on counterfactual scenarios. In particular, given a planning problem $P$ and an \ltlf formula $\psi$ defining desired properties of a plan, counterfactual scenarios identify minimal modifications to $P$ such that it admits plans that comply with $\psi$. In this paper, we present two qualitative instantiations of counterfactual scenarios based on an explicit quantification over plans that must satisfy $\psi$. We then characterise the computational complexity of generating such counterfactual scenarios when different types of changes are allowed on $P$. We show that producing counterfactual scenarios is often only as expensive as computing a plan for $P$, thus demonstrating the practical viability of our proposal and ultimately providing a framework to construct practical algorithms in this area.
comment: Accepted at the 22nd International Conference on Principles of Knowledge Representation and Reasoning (KR 2025)
Modeling Wise Decision Making: A Z-Number Fuzzy Framework Inspired by Phronesis
Background: Wisdom is a superordinate construct that embraces perspective taking, reflectiveness, prosocial orientation, reflective empathetic action, and intellectual humility. Unlike conventional models of reasoning that are rigidly bound by binary thinking, wisdom unfolds in shades of ambiguity, requiring both graded evaluation and self-reflective humility. Current measures depend on self-reports and seldom reflect the humility and uncertainty inherent in wise reasoning. A computational framework that takes into account both multidimensionality and confidence has the potential to improve psychological science and allow humane AI. Method: We present a fuzzy inference system with Z numbers, each of the decisions being expressed in terms of a wisdom score (restriction) and confidence score (certainty). As part of this study, participants (N = 100) were exposed to culturally neutral pictorial moral dilemma tasks to which they generated think-aloud linguistic responses, which were mapped into five theoretically based components of wisdom. The scores of each individual component were combined using a base of 21 rules, with membership functions tuned via Gaussian kernel density estimation. Results: In a proof of concept study, the system produced dual attribute wisdom representations that correlated modestly but significantly with established scales while showing negligible relations with unrelated traits, supporting convergent and divergent validity. Contribution: The contribution is to formalize wisdom as a multidimensional, uncertainty-conscious construct, operationalized in the form of Z-numbers. In addition to progressing measurement in psychology, it calculates how fuzzy Z numbers can provide AI systems with interpretable, confidence-sensitive reasoning that affords a safe, middle ground between rigorous computation and human-like judgment.
comment: total 17 pages, main manuscript 12 pages, supplementary 5 pages, 6 tables in main manuscript, 5 figures in main manuscript, 2 tables in supplementary, and 3 figures in supplementary
On the Hardness of Learning GNN-based SAT Solvers: The Role of Graph Ricci Curvature
Graph Neural Networks (GNNs) have recently shown promise as solvers for Boolean Satisfiability Problems (SATs) by operating on graph representations of logical formulas. However, their performance degrades sharply on harder instances, raising the question of whether this reflects fundamental architectural limitations. In this work, we provide a geometric explanation through the lens of graph Ricci Curvature (RC), which quantifies local connectivity bottlenecks. We prove that bipartite graphs derived from random k-SAT formulas are inherently negatively curved, and that this curvature decreases with instance difficulty. Building on this, we show that GNN-based SAT solvers are affected by oversquashing, a phenomenon where long-range dependencies become impossible to compress into fixed-length representations. We validate our claims empirically across different SAT benchmarks and confirm that curvature is both a strong indicator of problem complexity and can be used to predict performance. Finally, we connect our findings to design principles of existing solvers and outline promising directions for future work.
comment: Preprint
ELV-Halluc: Benchmarking Semantic Aggregation Hallucinations in Long Video Understanding
Video multimodal large language models (Video-MLLMs) have achieved remarkable progress in video understanding. However, they remain vulnerable to hallucination-producing content inconsistent with or unrelated to video inputs. Previous video hallucination benchmarks primarily focus on short-videos. They attribute hallucinations to factors such as strong language priors, missing frames, or vision-language biases introduced by the visual encoder. While these causes indeed account for most hallucinations in short videos, they still oversimplify the cause of hallucinations. Sometimes, models generate incorrect outputs but with correct frame-level semantics. We refer to this type of hallucination as Semantic Aggregation Hallucination (SAH), which arises during the process of aggregating frame-level semantics into event-level semantic groups. Given that SAH becomes particularly critical in long videos due to increased semantic complexity across multiple events, it is essential to separate and thoroughly investigate the causes of this type of hallucination. To address the above issues, we introduce ELV-Halluc, the first benchmark dedicated to long-video hallucination, enabling a systematic investigation of SAH. Our experiments confirm the existence of SAH and show that it increases with semantic complexity. Additionally, we find that models are more prone to SAH on rapidly changing semantics. Moreover, we discuss potential approaches to mitigate SAH. We demonstrate that positional encoding strategy contributes to alleviating SAH, and further adopt DPO strategy to enhance the model's ability to distinguish semantics within and across events. To support this, we curate a dataset of 8K adversarial data pairs and achieve improvements on both ELV-Halluc and Video-MME, including a substantial 27.7% reduction in SAH ratio.
Priors Matter: Addressing Misspecification in Bayesian Deep Q-Learning
Uncertainty quantification in reinforcement learning can greatly improve exploration and robustness. Approximate Bayesian approaches have recently been popularized to quantify uncertainty in model-free algorithms. However, so far the focus has been on improving the accuracy of the posterior approximation, instead of studying the accuracy of the prior and likelihood assumptions underlying the posterior. In this work, we demonstrate that there is a cold posterior effect in Bayesian deep Q-learning, where contrary to theory, performance increases when reducing the temperature of the posterior. To identify and overcome likely causes, we challenge common assumptions made on the likelihood and priors in Bayesian model-free algorithms. We empirically study prior distributions and show through statistical tests that the common Gaussian likelihood assumption is frequently violated. We argue that developing more suitable likelihoods and priors should be a key focus in future Bayesian reinforcement learning research and we offer simple, implementable solutions for better priors in deep Q-learning that lead to more performant Bayesian algorithms.
HSFN: Hierarchical Selection for Fake News Detection building Heterogeneous Ensemble
Psychological biases, such as confirmation bias, make individuals particularly vulnerable to believing and spreading fake news on social media, leading to significant consequences in domains such as public health and politics. Machine learning-based fact-checking systems have been widely studied to mitigate this problem. Among them, ensemble methods are particularly effective in combining multiple classifiers to improve robustness. However, their performance heavily depends on the diversity of the constituent classifiers-selecting genuinely diverse models remains a key challenge, especially when models tend to learn redundant patterns. In this work, we propose a novel automatic classifier selection approach that prioritizes diversity, also extended by performance. The method first computes pairwise diversity between classifiers and applies hierarchical clustering to organize them into groups at different levels of granularity. A HierarchySelect then explores these hierarchical levels to select one pool of classifiers per level, each representing a distinct intra-pool diversity. The most diverse pool is identified and selected for ensemble construction from these. The selection process incorporates an evaluation metric reflecting each classifiers's performance to ensure the ensemble also generalises well. We conduct experiments with 40 heterogeneous classifiers across six datasets from different application domains and with varying numbers of classes. Our method is compared against the Elbow heuristic and state-of-the-art baselines. Results show that our approach achieves the highest accuracy on two of six datasets. The implementation details are available on the project's repository: https://github.com/SaraBCoutinho/HSFN .
comment: Accepted by IEEE International Conference on Systems, Man, and Cybernetics (SMC) - IEEE SMC 2025
Igniting Creative Writing in Small Language Models: LLM-as-a-Judge versus Multi-Agent Refined Rewards EMNLP 2025
Large Language Models (LLMs) have demonstrated remarkable creative writing capabilities, yet their substantial computational demands hinder widespread use. Enhancing Small Language Models (SLMs) offers a promising alternative, but current methods like Supervised Fine-Tuning (SFT) struggle with novelty, and Reinforcement Learning from Human Feedback (RLHF) is costly. This paper explores two distinct AI-driven reward strategies within a Reinforcement Learning from AI Feedback (RLAIF) framework to ignite the creative writing of a 7B-parameter SLM, specifically for generating Chinese greetings. The first strategy employs a RM trained on high-quality preference data curated by a novel multi-agent rejection sampling framework designed for creative tasks. The second, more novel strategy utilizes a principle-guided LLM-as-a-Judge, whose reward function is optimized via an adversarial training scheme with a reflection mechanism, to directly provide reward signals. Comprehensive experiments reveal that while both approaches significantly enhance creative output over baselines, the principle-guided LLM-as-a-Judge demonstrably yields superior generation quality. Furthermore, it offers notable advantages in training efficiency and reduced dependency on human-annotated data, presenting a more scalable and effective path towards creative SLMs. Our automated evaluation methods also exhibit strong alignment with human judgments. Our code and data are publicly available at https://github.com/weixiaolong94-hub/Igniting-Creative-Writing-in-Small-Language-Models.
comment: EMNLP 2025 Main
MMSearch-Plus: A Simple Yet Challenging Benchmark for Multimodal Browsing Agents
Large multimodal language models (MLLMs) are increasingly deployed as web agents, yet many multimodal browsing benchmarks can be solved by shallow, fixed workflows that lean on high-recall image search and nearby text-masking the genuinely multimodal challenges of fine-grained visual reasoning, provenance verification, and long-horizon tool use. We introduce MMSearch-Plus, a benchmark of 311 tasks that highly demand multimodal understanding while preserving the difficulty profile of strong text-only browsing suites. Each item is constructed to contain multiple weak, localized visual signals that must be extracted, propagated through iterative text-image search, and cross-validated under retrieval noise before answering. Our curation procedure, Spatial-Temporal Extrapolation, seeds questions whose answers require extrapolating from spatial cues (micro-text, part-level appearance, layouts, signage) and temporal traces (broadcast overlays, seasonal context) to out-of-image facts such as events, dates, and venues. We provide a model-agnostic agent framework with browsing tools and evaluate a range of closed and open MLLMs. The strongest agent (o3) attains 15.1% without search and 36.0% accuracy with rollout under our framework, while a strong open-source model (Qwen-2.5-VL-72B-Instruct) achieves 0.0% without search and 6.9% after 20 rounds of search. Beyond answer accuracy, we assess bounding-box production and cropped-image search, and conduct an error analysis that surfaces failures in source verification, part-based reasoning, and long-horizon planning.
comment: Project Page: https://mmsearch-plus.github.io
Controllable 3D Molecular Generation for Structure-Based Drug Design Through Bayesian Flow Networks and Gradient Integration
Recent advances in Structure-based Drug Design (SBDD) have leveraged generative models for 3D molecular generation, predominantly evaluating model performance by binding affinity to target proteins. However, practical drug discovery necessitates high binding affinity along with synthetic feasibility and selectivity, critical properties that were largely neglected in previous evaluations. To address this gap, we identify fundamental limitations of conventional diffusion-based generative models in effectively guiding molecule generation toward these diverse pharmacological properties. We propose CByG, a novel framework extending Bayesian Flow Network into a gradient-based conditional generative model that robustly integrates property-specific guidance. Additionally, we introduce a comprehensive evaluation scheme incorporating practical benchmarks for binding affinity, synthetic feasibility, and selectivity, overcoming the limitations of conventional evaluation methods. Extensive experiments demonstrate that our proposed CByG framework significantly outperforms baseline models across multiple essential evaluation criteria, highlighting its effectiveness and practicality for real-world drug discovery applications.
Diffusion-based Multi-modal Synergy Interest Network for Click-through Rate Prediction SIGIR 2025
In click-through rate prediction, click-through rate prediction is used to model users' interests. However, most of the existing CTR prediction methods are mainly based on the ID modality. As a result, they are unable to comprehensively model users' multi-modal preferences. Therefore, it is necessary to introduce multi-modal CTR prediction. Although it seems appealing to directly apply the existing multi-modal fusion methods to click-through rate prediction models, these methods (1) fail to effectively disentangle commonalities and specificities across different modalities; (2) fail to consider the synergistic effects between modalities and model the complex interactions between modalities. To address the above issues, this paper proposes the Diffusion-based Multi-modal Synergy Interest Network (Diff-MSIN) framework for click-through prediction. This framework introduces three innovative modules: the Multi-modal Feature Enhancement (MFE) Module Synergistic Relationship Capture (SRC) Module, and the Feature Dynamic Adaptive Fusion (FDAF) Module. The MFE Module and SRC Module extract synergistic, common, and special information among different modalities. They effectively enhances the representation of the modalities, improving the overall quality of the fusion. To encourage distinctiveness among different features, we design a Knowledge Decoupling method. Additionally, the FDAF Module focuses on capturing user preferences and reducing fusion noise. To validate the effectiveness of the Diff-MSIN framework, we conducted extensive experiments using the Rec-Tmall and three Amazon datasets. The results demonstrate that our approach yields a significant improvement of at least 1.67% compared to the baseline, highlighting its potential for enhancing multi-modal recommendation systems. Our code is available at the following link: https://github.com/Cxx-0/Diff-MSIN.
comment: SIGIR 2025
Learning Lifted Action Models From Traces of Incomplete Actions and States KR 2025
Consider the problem of learning a lifted STRIPS model of the sliding-tile puzzle from random state-action traces where the states represent the location of the tiles only, and the actions are the labels up, down, left, and right, with no arguments. Two challenges are involved in this problem. First, the states are not full STRIPS states, as some predicates are missing, like the atoms representing the position of the ``blank''. Second, the actions are not full STRIPS either, as they do not reveal all the objects involved in the actions effects and preconditions. Previous approaches have addressed different versions of this model learning problem, but most assume that actions in the traces are full STRIPS actions or that the domain predicates are all observable. The new setting considered in this work is more ``realistic'', as the atoms observed convey the state of the world but not full STRIPS states, and the actions reveal the arguments needed for selecting the action but not the ones needed for modeling it in STRIPS. For formulating and addressing the learning problem, we introduce a variant of STRIPS, which we call STRIPS+, where certain STRIPS action arguments can be left implicit in preconditions which can also involve a limited form of existential quantification. The learning problem becomes the problem of learning STRIPS+ models from STRIPS+ state-action traces. For this, the proposed learning algorithm, called SYNTH, constructs a stratified sequence (conjunction) of precondition expressions or ``queries'' for each action, that denote unique objects in the state and ground the implicit action arguments in STRIPS+. The correctness and completeness of SYNTH is established, and its scalability is tested on state-action traces obtained from STRIPS+ models derived from existing STRIPS domains.
comment: To be presented at KR 2025
A General Framework of Epistemic Forgetting and its Instantiation by Ranking Functions
Forgetting as a knowledge management operation deliberately ignores parts of the knowledge and beliefs of an agent, for various reasons. Forgetting has many facets, one may want to forget parts of the syntax, a proposition, or a conditional. In the literature, two main operators suitable for performing forgetting have been proposed and investigated in depth: First, variable elimination is a syntactical method that blends out certain atomic variables to focus on the rest of the language. It has been mainly used in the area of logic programming and answer set programming. Second, contraction in AGM belief revision theory effectively removes propositions from belief sets under logical deduction. Both operations rely mainly on classical logics. In this article, we take an epistemic perspective and study forgetting operations in epistemic states with richer semantic structures, but with clear links to propositional logic. This allows us to investigate what forgetting in the epistemic background means, thereby lifting well-known and novel forgetting operations to the epistemic level. We present five general types of epistemic forgetting and instantiate them with seven concrete forgetting operations for Spohn's ranking functions. We take inspiration from postulates of forgetting both from logic programming and AGM theory to propose a rich landscape of axioms for evaluating forgetting operations. Finally, we evaluate all concrete forgetting operations according to all postulates, leading to a novel comprehensive overview highlighting differences and commonalities among the forgetting operators.
MedShift: Implicit Conditional Transport for X-Ray Domain Adaptation ICCV 2025
Synthetic medical data offers a scalable solution for training robust models, but significant domain gaps limit its generalizability to real-world clinical settings. This paper addresses the challenge of cross-domain translation between synthetic and real X-ray images of the head, focusing on bridging discrepancies in attenuation behavior, noise characteristics, and soft tissue representation. We propose MedShift, a unified class-conditional generative model based on Flow Matching and Schrodinger Bridges, which enables high-fidelity, unpaired image translation across multiple domains. Unlike prior approaches that require domain-specific training or rely on paired data, MedShift learns a shared domain-agnostic latent space and supports seamless translation between any pair of domains seen during training. We introduce X-DigiSkull, a new dataset comprising aligned synthetic and real skull X-rays under varying radiation doses, to benchmark domain translation models. Experimental results demonstrate that, despite its smaller model size compared to diffusion-based approaches, MedShift offers strong performance and remains flexible at inference time, as it can be tuned to prioritize either perceptual fidelity or structural consistency, making it a scalable and generalizable solution for domain adaptation in medical imaging. The code and dataset are available at https://caetas.github.io/medshift.html
comment: Accepted at the ICCV 2025 AIM Workshop
The Complexity Trap: Simple Observation Masking Is as Efficient as LLM Summarization for Agent Context Management
Large Language Model (LLM)-based agents solve complex tasks through iterative reasoning, exploration, and tool-use, a process that can result in long, expensive context histories. While state-of-the-art Software Engineering ( SE) agents like OpenHands or Cursor use LLM-based summarization to tackle this issue, it is unclear whether the increased complexity offers tangible performance benefits compared to simply omitting older observations. We present a systematic comparison of these strategies within SWE-agent on SWE-bench Verified across five diverse model configurations. We find that a simple observation-masking strategy halves cost relative to a raw agent while matching, and sometimes slightly exceeding, the solve rate of LLM summarization. For example, with Qwen3-Coder 480B, masking improves solve rate from 53.8% (raw agent) to 54.8%, while remaining competitive with summarization at a lower cost. These results suggest that, at least within SWE-agent on SWE-bench Verified, the most effective and efficient context management can be the simplest. We release code and data for reproducibility
Med-RewardBench: Benchmarking Reward Models and Judges for Medical Multimodal Large Language Models
Multimodal large language models (MLLMs) hold significant potential in medical applications, including disease diagnosis and clinical decision-making. However, these tasks require highly accurate, context-sensitive, and professionally aligned responses, making reliable reward models and judges critical. Despite their importance, medical reward models (MRMs) and judges remain underexplored, with no dedicated benchmarks addressing clinical requirements. Existing benchmarks focus on general MLLM capabilities or evaluate models as solvers, neglecting essential evaluation dimensions like diagnostic accuracy and clinical relevance. To address this, we introduce Med-RewardBench, the first benchmark specifically designed to evaluate MRMs and judges in medical scenarios. Med-RewardBench features a multimodal dataset spanning 13 organ systems and 8 clinical departments, with 1,026 expert-annotated cases. A rigorous three-step process ensures high-quality evaluation data across six clinically critical dimensions. We evaluate 32 state-of-the-art MLLMs, including open-source, proprietary, and medical-specific models, revealing substantial challenges in aligning outputs with expert judgment. Additionally, we develop baseline models that demonstrate substantial performance improvements through fine-tuning.
comment: 19 pages, 5 figures, 3 tables
Benchmarking the State of Networks with a Low-Cost Method Based on Reservoir Computing
Using data from mobile network utilization in Norway, we showcase the possibility of monitoring the state of communication and mobility networks with a non-invasive, low-cost method. This method transforms the network data into a model within the framework of reservoir computing and then measures the model's performance on proxy tasks. Experimentally, we show how the performance on these proxies relates to the state of the network. A key advantage of this approach is that it uses readily available data sets and leverages the reservoir computing framework for an inexpensive and largely agnostic method. Data from mobile network utilization is available in an anonymous, aggregated form with multiple snapshots per day. This data can be treated like a weighted network. Reservoir computing allows the use of weighted, but untrained networks as a machine learning tool. The network, initialized as a so-called echo state network (ESN), projects incoming signals into a higher dimensional space, on which a single trained layer operates. This consumes less energy than deep neural networks in which every weight of the network is trained. We use neuroscience inspired tasks and trained our ESN model to solve them. We then show how the performance depends on certain network configurations and also how it visibly decreases when perturbing the network. While this work serves as proof of concept, we believe it can be elevated to be used for near-real-time monitoring as well as the identification of possible weak spots of both mobile communication networks as well as transportation networks.
comment: Net-Zero Future 2025 Conference
CARJAN: Agent-Based Generation and Simulation of Traffic Scenarios with AJAN
User-friendly modeling and virtual simulation of urban traffic scenarios with different types of interacting agents such as pedestrians, cyclists and autonomous vehicles remains a challenge. We present CARJAN, a novel tool for semi-automated generation and simulation of such scenarios based on the multi-agent engineering framework AJAN and the driving simulator CARLA. CARJAN provides a visual user interface for the modeling, storage and maintenance of traffic scenario layouts, and leverages SPARQL Behavior Tree-based decision-making and interactions for agents in dynamic scenario simulations in CARLA. CARJAN provides a first integrated approach for interactive, intelligent agent-based generation and simulation of virtual traffic scenarios in CARLA.
DRASP: A Dual-Resolution Attentive Statistics Pooling Framework for Automatic MOS Prediction SC 2025
A pooling mechanism is essential for mean opinion score (MOS) prediction, facilitating the transformation of variable-length audio features into a concise fixed-size representation that effectively encodes speech quality. Existing pooling methods typically operate at a singular granularity, concentrating either on a comprehensive global perspective or a detailed frame-level analysis, which may overlook complementary perceptual insights. To address this limitation, we introduce the Dual-Resolution Attentive Statistics Pooling (DRASP) framework. DRASP integrates both coarse-grained, global statistical summaries and fine-grained, attentive analyses of perceptually significant segments. This dual-view architecture empowers our model to formulate a more thorough and robust representation, capturing both the overarching structural context and salient local details concurrently. Extensive experiments validate the effectiveness and strong generalization ability of the proposed framework. It consistently outperforms various baseline methods across diverse datasets (MusicEval and AES-Natural), MOS prediction backbones (including a CLAP-based model and AudioBox-Aesthetics), and different audio generation systems, achieving a relative improvement of 10.39% in system-level Spearman's rank correlation coefficient (SRCC) over the widely-used average pooling approach.
comment: Accepted to APSIPA ASC 2025
AI Compute Architecture and Evolution Trends
The focus of AI development has shifted from academic research to practical applications. However, AI development faces numerous challenges at various levels. This article will attempt to analyze the opportunities and challenges of AI from several different perspectives using a structured approach. This article proposes a seven-layer model for AI compute architecture, including Physical Layer, Link Layer, Neural Network Layer, Context Layer, Agent Layer, Orchestrator Layer, and Application Layer, from bottom to top. It also explains how AI computing has evolved into this 7-layer architecture through the three-stage evolution on large-scale language models (LLMs). For each layer, we describe the development trajectory and key technologies. In Layers 1 and 2 we discuss AI computing issues and the impact of Scale-Up and Scale-Out strategies on computing architecture. In Layer 3 we explore two different development paths for LLMs. In Layer 4 we discuss the impact of contextual memory on LLMs and compares it to traditional processor memory. In Layers 5 to 7 we discuss the trends of AI agents and explore the issues in evolution from a single AI agent to an AI-based ecosystem, and their impact on the AI industry. Furthermore, AI development involves not only technical challenges but also the economic issues to build self-sustainable ecosystem. This article analyzes the internet industry to provide predictions on the future trajectory of AI development.
comment: 29 pages, 26 figures
zkLoRA: Fine-Tuning Large Language Models with Verifiable Security via Zero-Knowledge Proofs
Fine-tuning large language models (LLMs) is crucial for adapting them to specific tasks, yet it remains computationally demanding and raises concerns about correctness and privacy, particularly in untrusted environments. Although parameter-efficient methods like Low-Rank Adaptation (LoRA) significantly reduce resource requirements, ensuring the security and verifiability of fine-tuning under zero-knowledge constraints remains an unresolved challenge. To address this, we introduce zkLoRA, the first framework to integrate LoRA fine-tuning with zero-knowledge proofs (ZKPs), achieving provable security and correctness. zkLoRA employs advanced cryptographic techniques -- such as lookup arguments, sumcheck protocols, and polynomial commitments -- to verify both arithmetic and non-arithmetic operations in Transformer-based architectures. The framework provides end-to-end verifiability for forward propagation, backward propagation, and parameter updates during LoRA fine-tuning, while safeguarding the privacy of model parameters and training data. Leveraging GPU-based implementations, zkLoRA demonstrates practicality and efficiency through experimental validation on open-source LLMs like LLaMA, scaling up to 13 billion parameters. By combining parameter-efficient fine-tuning with ZKPs, zkLoRA bridges a critical gap, enabling secure and trustworthy deployment of LLMs in sensitive or untrusted environments.
AllSummedUp: un framework open-source pour comparer les metriques d'evaluation de resume
This paper investigates reproducibility challenges in automatic text summarization evaluation. Based on experiments conducted across six representative metrics ranging from classical approaches like ROUGE to recent LLM-based methods (G-Eval, SEval-Ex), we highlight significant discrepancies between reported performances in the literature and those observed in our experimental setting. We introduce a unified, open-source framework, applied to the SummEval dataset and designed to support fair and transparent comparison of evaluation metrics. Our results reveal a structural trade-off: metrics with the highest alignment with human judgments tend to be computationally intensive and less stable across runs. Beyond comparative analysis, this study highlights key concerns about relying on LLMs for evaluation, stressing their randomness, technical dependencies, and limited reproducibility. We advocate for more robust evaluation protocols including exhaustive documentation and methodological standardization to ensure greater reliability in automatic summarization assessment.
comment: in French language
Normality and the Turing Test
This paper proposes to revisit the Turing test through the concept of normality. Its core argument is that the statistical interpretation of the normal--understood as the average both in the normative and mathematical sense of the term--proves useful for understanding the Turing test in at least two ways. First, in the sense that the Turing test targets normal/average rather than exceptional human intelligence, so that successfully passing the test requires building machines that "make mistakes" and display imperfect behavior just like normal/average humans. Second, in the sense that the Turing test is a statistical test where judgments of intelligence are never carried out by a single "average" judge (understood as non-expert) but always by a full jury. As such, the notion of "average human interrogator" that Turing talks about in his original paper should be understood primarily as referring to a mathematical abstraction made of the normalized aggregate of individual judgments of multiple judges. In short, this paper argues that the Turing test is a test of normal intelligence as assessed by a normal judge characterizing the average judgment of a pool of human interrogators. Its conclusions are twofold. First, it argues that large language models such as ChatGPT are unlikely to pass the Turing test as those models precisely target exceptional rather than normal/average human intelligence. As such, they constitute models of what it proposes to call artificial smartness rather than artificial intelligence per se. Second, it argues that the core question of whether the Turing test can contribute anything to the understanding of human cognition is that of whether the human mind is really reducible to the normal/average mind--a question which largely extends beyond the Turing test itself and questions the conceptual underpinnings of the normalist paradigm it belongs to.
Iterative Inference in a Chess-Playing Neural Network
Do neural networks build their representations through smooth, gradual refinement, or via more complex computational processes? We investigate this by extending the logit lens to analyze the policy network of Leela Chess Zero, a superhuman chess engine. We find strong monotonic trends in playing strength and puzzle-solving ability across layers, yet policy distributions frequently follow non-smooth trajectories. Evidence for this includes correct puzzle solutions that are discovered early but subsequently discarded, move rankings that remain poorly correlated with final outputs, and high policy divergence until late in the network. These findings contrast with the smooth distributional convergence typically observed in language models.
RoboInspector: Unveiling the Unreliability of Policy Code for LLM-enabled Robotic Manipulation
Large language models (LLMs) demonstrate remarkable capabilities in reasoning and code generation, enabling robotic manipulation to be initiated with just a single instruction. The LLM carries out various tasks by generating policy code required to control the robot. Despite advances in LLMs, achieving reliable policy code generation remains a significant challenge due to the diverse requirements of real-world tasks and the inherent complexity of user instructions. In practice, different users may provide distinct instructions to drive the robot for the same task, which may cause the unreliability of policy code generation. To bridge this gap, we design RoboInspector, a pipeline to unveil and characterize the unreliability of the policy code for LLM-enabled robotic manipulation from two perspectives: the complexity of the manipulation task and the granularity of the instruction. We perform comprehensive experiments with 168 distinct combinations of tasks, instructions, and LLMs in two prominent frameworks. The RoboInspector identifies four main unreliable behaviors that lead to manipulation failure. We provide a detailed characterization of these behaviors and their underlying causes, giving insight for practical development to reduce unreliability. Furthermore, we introduce a refinement approach guided by failure policy code feedback that improves the reliability of policy code generation by up to 35% in LLM-enabled robotic manipulation, evaluated in both simulation and real-world environments.
Challenges and Applications of Large Language Models: A Comparison of GPT and DeepSeek family of models
Large Language Models (LLMs) are transforming AI across industries, but their development and deployment remain complex. This survey reviews 16 key challenges in building and using LLMs and examines how these challenges are addressed by two state-of-the-art models with unique approaches: OpenAI's closed source GPT-4o (May 2024 update) and DeepSeek-V3-0324 (March 2025), a large open source Mixture-of-Experts model. Through this comparison, we showcase the trade-offs between closed source models (robust safety, fine-tuned reliability) and open source models (efficiency, adaptability). We also explore LLM applications across different domains (from chatbots and coding tools to healthcare and education), highlighting which model attributes are best suited for each use case. This article aims to guide AI researchers, developers, and decision-makers in understanding current LLM capabilities, limitations, and best practices.
comment: 18 pages, 7 figures
AHELM: A Holistic Evaluation of Audio-Language Models
Evaluations of audio-language models (ALMs) -- multimodal models that take interleaved audio and text as input and output text -- are hindered by the lack of standardized benchmarks; most benchmarks measure only one or two capabilities and omit evaluative aspects such as fairness or safety. Furthermore, comparison across models is difficult as separate evaluations test a limited number of models and use different prompting methods and inference parameters. To address these shortfalls, we introduce AHELM, a benchmark that aggregates various datasets -- including 2 new synthetic audio-text datasets called PARADE, which evaluates the ALMs on avoiding stereotypes, and CoRe-Bench, which measures reasoning over conversational audio through inferential multi-turn question answering -- to holistically measure the performance of ALMs across 10 aspects we have identified as important to the development and usage of ALMs: audio perception, knowledge, reasoning, emotion detection, bias, fairness, multilinguality, robustness, toxicity, and safety. We also standardize the prompts, inference parameters, and evaluation metrics to ensure equitable comparisons across models. We test 14 open-weight and closed-API ALMs from 3 developers and 3 additional simple baseline systems each consisting of an automatic speech recognizer and a language model. Our results show that while Gemini 2.5 Pro ranks top in 5 out of 10 aspects, it exhibits group unfairness ($p=0.01$) on ASR tasks whereas most of the other models do not. We also find that the baseline systems perform reasonably well on AHELM, with one ranking 5th overall despite having only speech-to-text capabilities. For transparency, all raw prompts, model generations, and outputs are available on our website at https://crfm.stanford.edu/helm/audio/v1.0.0. AHELM is intended to be a living benchmark and new datasets and models will be added over time.
EconAgentic in DePIN Markets: A Large Language Model Approach to the Sharing Economy of Decentralized Physical Infrastructure
The Decentralized Physical Infrastructure (DePIN) market is revolutionizing the sharing economy through token-based economics and smart contracts that govern decentralized operations. By 2024, DePIN projects have exceeded \$10 billion in market capitalization, underscoring their rapid growth. However, the unregulated nature of these markets, coupled with the autonomous deployment of AI agents in smart contracts, introduces risks such as inefficiencies and potential misalignment with human values. To address these concerns, we introduce EconAgentic, a Large Language Model (LLM)-powered framework designed to mitigate these challenges. Our research focuses on three key areas: 1) modeling the dynamic evolution of DePIN markets, 2) evaluating stakeholders' actions and their economic impacts, and 3) analyzing macroeconomic indicators to align market outcomes with societal goals. Through EconAgentic, we simulate how AI agents respond to token incentives, invest in infrastructure, and adapt to market conditions, comparing AI-driven decisions with human heuristic benchmarks. Our results show that EconAgentic provides valuable insights into the efficiency, inclusion, and stability of DePIN markets, contributing to both academic understanding and practical improvements in the design and governance of decentralized, tokenized economies.
Think in Games: Learning to Reason in Games via Reinforcement Learning with Large Language Models
Large language models (LLMs) excel at complex reasoning tasks such as mathematics and coding, yet they frequently struggle with simple interactive tasks that young children perform effortlessly. This discrepancy highlights a critical gap between declarative knowledge (knowing about something) and procedural knowledge (knowing how to do something). Although traditional reinforcement learning (RL) agents can acquire procedural knowledge through environmental interaction, they often operate as black boxes and require substantial training data. In contrast, LLMs possess extensive world knowledge and reasoning capabilities, but are unable to effectively convert this static knowledge into dynamic decision-making in interactive settings. To address this challenge, we propose Think in Games (TiG), a novel framework that empowers LLMs to develop procedural understanding through direct interaction with game environments, while retaining their inherent reasoning and explanatory abilities. Specifically, TiG reformulates RL-based decision-making as a language modeling task: LLMs generate language-guided policies, which are refined iteratively through online reinforcement learning based on environmental feedback. Our experimental results show that TiG successfully bridges the gap between declarative and procedural knowledge, achieving competitive performance with dramatically lower data and computational demands compared to conventional RL methods. Moreover, TiG provides step-by-step natural language explanations for its decisions, greatly improving transparency and interpretability in complex interactive tasks.
Adaptive Heavy-Tailed Stochastic Gradient Descent
In the era of large-scale neural network models, optimization algorithms often struggle with generalization due to an overreliance on training loss. One key insight widely accepted in the machine learning community is the idea that wide basins (regions around a local minimum where the loss increases gradually) promote better generalization by offering greater stability to small changes in input data or model parameters. In contrast, sharp minima are typically more sensitive and less stable. Motivated by two key empirical observations - the inherent heavy-tailed distribution of gradient noise in stochastic gradient descent and the Edge of Stability phenomenon during neural network training, in which curvature grows before settling at a plateau, we introduce Adaptive Heavy Tailed Stochastic Gradient Descent (AHTSGD). The algorithm injects heavier-tailed noise into the optimizer during the early stages of training to enhance exploration and gradually transitions to lighter-tailed noise as sharpness stabilizes. By dynamically adapting to the sharpness of the loss landscape throughout training, AHTSGD promotes accelerated convergence to wide basins. AHTSGD is the first algorithm to adjust the nature of injected noise into an optimizer based on the Edge of Stability phenomenon. AHTSGD consistently outperforms SGD and other noise-based methods on benchmarks like MNIST and CIFAR-10, with marked gains on noisy datasets such as SVHN. It ultimately accelerates early training from poor initializations and improves generalization across clean and noisy settings, remaining robust to learning rate choices.
DLGAN : Time Series Synthesis Based on Dual-Layer Generative Adversarial Networks
Time series synthesis is an effective approach to ensuring the secure circulation of time series data. Existing time series synthesis methods typically perform temporal modeling based on random sequences to generate target sequences, which often struggle to ensure the temporal dependencies in the generated time series. Additionally, directly modeling temporal features on random sequences makes it challenging to accurately capture the feature information of the original time series. To address the above issues, we propose a simple but effective generative model \textbf{D}ual-\textbf{L}ayer \textbf{G}enerative \textbf{A}dversarial \textbf{N}etworks, named \textbf{DLGAN}. The model decomposes the time series generation process into two stages: sequence feature extraction and sequence reconstruction. First, these two stages form a complete time series autoencoder, enabling supervised learning on the original time series to ensure that the reconstruction process can restore the temporal dependencies of the sequence. Second, a Generative Adversarial Network (GAN) is used to generate synthetic feature vectors that align with the real-time sequence feature vectors, ensuring that the generator can capture the temporal features from real time series. Extensive experiments on four public datasets demonstrate the superiority of this model across various evaluation metrics.
comment: 8 pages, 3 figures
Stairway to Fairness: Connecting Group and Individual Fairness RecSys 2025
Fairness in recommender systems (RSs) is commonly categorised into group fairness and individual fairness. However, there is no established scientific understanding of the relationship between the two fairness types, as prior work on both types has used different evaluation measures or evaluation objectives for each fairness type, thereby not allowing for a proper comparison of the two. As a result, it is currently not known how increasing one type of fairness may affect the other. To fill this gap, we study the relationship of group and individual fairness through a comprehensive comparison of evaluation measures that can be used for both fairness types. Our experiments with 8 runs across 3 datasets show that recommendations that are highly fair for groups can be very unfair for individuals. Our finding is novel and useful for RS practitioners aiming to improve the fairness of their systems. Our code is available at: https://github.com/theresiavr/stairway-to-fairness.
comment: Accepted to RecSys 2025 (short paper)
Stage-Diff: Stage-wise Long-Term Time Series Generation Based on Diffusion Models
Generative models have been successfully used in the field of time series generation. However, when dealing with long-term time series, which span over extended periods and exhibit more complex long-term temporal patterns, the task of generation becomes significantly more challenging. Long-term time series exhibit long-range temporal dependencies, but their data distribution also undergoes gradual changes over time. Finding a balance between these long-term dependencies and the drift in data distribution is a key challenge. On the other hand, long-term time series contain more complex interrelationships between different feature sequences, making the task of effectively capturing both intra-sequence and inter-sequence dependencies another important challenge. To address these issues, we propose Stage-Diff, a staged generative model for long-term time series based on diffusion models. First, through stage-wise sequence generation and inter-stage information transfer, the model preserves long-term sequence dependencies while enabling the modeling of data distribution shifts. Second, within each stage, progressive sequence decomposition is applied to perform channel-independent modeling at different time scales, while inter-stage information transfer utilizes multi-channel fusion modeling. This approach combines the robustness of channel-independent modeling with the information fusion advantages of multi-channel modeling, effectively balancing the intra-sequence and inter-sequence dependencies of long-term time series. Extensive experiments on multiple real-world datasets validate the effectiveness of Stage-Diff in long-term time series generation tasks.
comment: 8 pages, 5 figures
Multi-Ontology Integration with Dual-Axis Propagation for Medical Concept Representation CIKM 2025
Medical ontology graphs map external knowledge to medical codes in electronic health records via structured relationships. By leveraging domain-approved connections (e.g., parent-child), predictive models can generate richer medical concept representations by incorporating contextual information from related concepts. However, existing literature primarily focuses on incorporating domain knowledge from a single ontology system, or from multiple ontology systems (e.g., diseases, drugs, and procedures) in isolation, without integrating them into a unified learning structure. Consequently, concept representation learning often remains limited to intra-ontology relationships, overlooking cross-ontology connections. In this paper, we propose LINKO, a large language model (LLM)-augmented integrative ontology learning framework that leverages multiple ontology graphs simultaneously by enabling dual-axis knowledge propagation both within and across heterogeneous ontology systems to enhance medical concept representation learning. Specifically, LINKO first employs LLMs to provide a graph-retrieval-augmented initialization for ontology concept embedding, through an engineered prompt that includes concept descriptions, and is further augmented with ontology context. Second, our method jointly learns the medical concepts in diverse ontology graphs by performing knowledge propagation in two axes: (1) intra-ontology vertical propagation across hierarchical ontology levels and (2) inter-ontology horizontal propagation within every level in parallel. Last, through extensive experiments on two public datasets, we validate the superior performance of LINKO over state-of-the-art baselines. As a plug-in encoder compatible with existing EHR predictive models, LINKO further demonstrates enhanced robustness in scenarios involving limited data availability and rare disease prediction.
comment: This work has been accepted as a full research paper at CIKM 2025
MultiFluxAI Enhancing Platform Engineering with Advanced Agent-Orchestrated Retrieval Systems
MultiFluxAI is an innovative AI platform developed to address the challenges of managing and integrating vast, disparate data sources in product engineering across application domains. It addresses both current and new service related queries that enhance user engagement in the digital ecosystem. This platform leverages advanced AI techniques, such as Generative AI, vectorization, and agentic orchestration to provide dynamic and context-aware responses to complex user queries.
comment: Abstract accepted for presentation at ACM ISEC 2025
Locus: Agentic Predicate Synthesis for Directed Fuzzing
Directed fuzzing aims to find program inputs that lead to specified target program states. It has broad applications, such as debugging system crashes, confirming reported bugs, and generating exploits for potential vulnerabilities. This task is inherently challenging because target states are often deeply nested in the program, while the search space manifested by numerous possible program inputs is prohibitively large. Existing approaches rely on branch distances or manually-specified constraints to guide the search; however, the branches alone are often insufficient to precisely characterize progress toward reaching the target states, while the manually specified constraints are often tailored for specific bug types and thus difficult to generalize to diverse target states and programs. We present Locus, a novel framework to improve the efficiency of directed fuzzing. Our key insight is to synthesize predicates to capture fuzzing progress as semantically meaningful intermediate states, serving as milestones towards reaching the target states. When used to instrument the program under fuzzing, they can reject executions unlikely to reach the target states, while providing additional coverage guidance. To automate this task and generalize to diverse programs, Locus features an agentic framework with program analysis tools to synthesize and iteratively refine the candidate predicates, while ensuring the predicates strictly relax the target states to prevent false rejections via symbolic execution. Our evaluation shows that Locus substantially improves the efficiency of eight state-of-the-art fuzzers in discovering real-world vulnerabilities, achieving an average speedup of 41.6x. So far, Locus has found eight previously unpatched bugs, with one already acknowledged with a draft patch.
MyGO: Memory Yielding Generative Offline-consolidation for Lifelong Learning Systems
Continual or Lifelong Learning aims to develop models capable of acquiring new knowledge from a sequence of tasks without catastrophically forgetting what has been learned before. Existing approaches often rely on storing samples from previous tasks (experience replay) or employing complex regularization terms to protect learned weights. However, these methods face challenges related to data privacy, storage limitations, and performance degradation when tasks are dissimilar. To address these challenges, we introduce MyGO (Memory Yielding Generative Offline-consolidation), a novel lifelong learning framework inspired by the biological wake-sleep cycle. During the "wake" phase, the system rapidly learns a new task and trains a compact generative model (Generative Memory, G-mem) to capture its data distribution. During the "sleep" phase, the system enters an offline state, using all learned G-mem models to generate pseudo-data ("dreams") and consolidate new and old knowledge into a core feature extractor via knowledge distillation. This approach obviates the need to store any raw data, retaining only compact generative models, which offers significant advantages in privacy and storage efficiency. We evaluate MyGO on computer vision (Split-MNIST) and natural language processing (Split-AG News) benchmarks, comparing it against a sequential fine-tuning baseline. The results demonstrate that MyGO significantly mitigates catastrophic forgetting and maintains high average accuracy across tasks, proving the framework's effectiveness and domain-generality.
comment: 5 pages
BLUEX Revisited: Enhancing Benchmark Coverage with Automatic Captioning
With the growing capabilities of Large Language Models (LLMs), there is an increasing need for robust evaluation methods, especially in multilingual and non-English contexts. We present an updated version of the BLUEX dataset, now including 2024-2025 exams and automatically generated image captions using state-of-the-art models, enhancing its relevance for data contamination studies in LLM pretraining. Captioning strategies increase accessibility to text-only models by more than 40%, producing 1,422 usable questions, more than doubling the number in the original BLUEX. We evaluated commercial and open-source LLMs and their ability to leverage visual context through captions.
comment: 12 pages, 5 figures, 2 tables
Efficient Code Embeddings from Code Generation Models
jina-code-embeddings is a novel code embedding model suite designed to retrieve code from natural language queries, perform technical question-answering, and identify semantically similar code snippets across programming languages. It makes innovative use of an autoregressive backbone pre-trained on both text and code, generating embeddings via last-token pooling. We outline the training recipe and demonstrate state-of-the-art performance despite the relatively small size of the models, validating this approach to code embedding model construction.
comment: 9 pages, table and evaluations 5-9
A Financial Brain Scan of the LLM
Emerging techniques in computer science make it possible to "brain scan" large language models (LLMs), identify the plain-English concepts that guide their reasoning, and steer them while holding other factors constant. We show that this approach can map LLM-generated economic forecasts to concepts such as sentiment, technical analysis, and timing, and compute their relative importance without reducing performance. We also show that models can be steered to be more or less risk-averse, optimistic, or pessimistic, which allows researchers to correct or simulate biases. The method is transparent, lightweight, and replicable for empirical research in the social sciences.
comment: 47 pages
SAGA: A Security Architecture for Governing AI Agentic Systems
Large Language Model (LLM)-based agents increasingly interact, collaborate, and delegate tasks to one another autonomously with minimal human interaction. Industry guidelines for agentic system governance emphasize the need for users to maintain comprehensive control over their agents, mitigating potential damage from malicious agents. Several proposed agentic system designs address agent identity, authorization, and delegation, but remain purely theoretical, without concrete implementation and evaluation. Most importantly, they do not provide user-controlled agent management. To address this gap, we propose SAGA, a scalable Security Architecture for Governing Agentic systems, that offers user oversight over their agents' lifecycle. In our design, users register their agents with a central entity, the Provider, that maintains agent contact information, user-defined access control policies, and helps agents enforce these policies on inter-agent communication. We introduce a cryptographic mechanism for deriving access control tokens, that offers fine-grained control over an agent's interaction with other agents, providing formal security guarantees. We evaluate SAGA on several agentic tasks, using agents in different geolocations, and multiple on-device and cloud LLMs, demonstrating minimal performance overhead with no impact on underlying task utility in a wide range of conditions. Our architecture enables secure and trustworthy deployment of autonomous agents, accelerating the responsible adoption of this technology in sensitive environments.
ROSE: A Reward-Oriented Data Selection Framework for LLM Task-Specific Instruction Tuning EMNLP 2025
Instruction tuning has underscored the significant potential of large language models (LLMs) in producing more human controllable and effective outputs in various domains. In this work, we focus on the data selection problem for task-specific instruction tuning of LLMs. Prevailing methods primarily rely on the crafted similarity metrics to select training data that aligns with the test data distribution. The goal is to minimize instruction tuning loss on the test data, ultimately improving performance on the target task. However, it has been widely observed that instruction tuning loss (i.e., cross-entropy loss for next token prediction) in LLMs often fails to exhibit a monotonic relationship with actual task performance. This misalignment undermines the effectiveness of current data selection methods for task-specific instruction tuning. To address this issue, we introduce ROSE, a novel Reward-Oriented inStruction data sElection method which leverages pairwise preference loss as a reward signal to optimize data selection for task-specific instruction tuning. Specifically, ROSE adapts an influence formulation to approximate the influence of training data points relative to a few-shot preference validation set to select the most task-related training data points. Experimental results show that by selecting just 5\% of the training data using ROSE, our approach can achieve competitive results compared to fine-tuning with the full training dataset, and it surpasses other state-of-the-art data selection methods for task-specific instruction tuning. Our qualitative analysis further confirms the robust generalizability of our method across multiple benchmark datasets and diverse model architectures.
comment: EMNLP 2025 Findings
Single Domain Generalization for Multimodal Cross-Cancer Prognosis via Dirac Rebalancer and Distribution Entanglement
Deep learning has shown remarkable performance in integrating multimodal data for survival prediction. However, existing multimodal methods mainly focus on single cancer types and overlook the challenge of generalization across cancers. In this work, we are the first to reveal that multimodal prognosis models often generalize worse than unimodal ones in cross-cancer scenarios, despite the critical need for such robustness in clinical practice. To address this, we propose a new task: Cross-Cancer Single Domain Generalization for Multimodal Prognosis, which evaluates whether models trained on a single cancer type can generalize to unseen cancers. We identify two key challenges: degraded features from weaker modalities and ineffective multimodal integration. To tackle these, we introduce two plug-and-play modules: Sparse Dirac Information Rebalancer (SDIR) and Cancer-aware Distribution Entanglement (CADE). SDIR mitigates the dominance of strong features by applying Bernoulli-based sparsification and Dirac-inspired stabilization to enhance weaker modality signals. CADE, designed to synthesize the target domain distribution, fuses local morphological cues and global gene expression in latent space. Experiments on a four-cancer-type benchmark demonstrate superior generalization, laying the foundation for practical, robust cross-cancer multimodal prognosis. Code is available at https://github.com/HopkinsKwong/MCCSDG
comment: Accepted by ACMMM 25
COBRA-PPM: A Causal Bayesian Reasoning Architecture Using Probabilistic Programming for Robot Manipulation Under Uncertainty
Manipulation tasks require robots to reason about cause and effect when interacting with objects. Yet, many data-driven approaches lack causal semantics and thus only consider correlations. We introduce COBRA-PPM, a novel causal Bayesian reasoning architecture that combines causal Bayesian networks and probabilistic programming to perform interventional inference for robot manipulation under uncertainty. We demonstrate its capabilities through high-fidelity Gazebo-based experiments on an exemplar block stacking task, where it predicts manipulation outcomes with high accuracy (Pred Acc: 88.6%) and performs greedy next-best action selection with a 94.2% task success rate. We further demonstrate sim2real transfer on a domestic robot, showing effectiveness in handling real-world uncertainty from sensor noise and stochastic actions. Our generalised and extensible framework supports a wide range of manipulation scenarios and lays a foundation for future work at the intersection of robotics and causality.
comment: 8 pages, 7 figures, accepted to the 2025 IEEE European Conference on Mobile Robots (ECMR 2025)
What Breaks Knowledge Graph based RAG? Empirical Insights into Reasoning under Incomplete Knowledge
Knowledge Graph-based Retrieval-Augmented Generation (KG-RAG) is an increasingly explored approach for combining the reasoning capabilities of large language models with the structured evidence of knowledge graphs. However, current evaluation practices fall short: existing benchmarks often include questions that can be directly answered using existing triples in KG, making it unclear whether models perform reasoning or simply retrieve answers directly. Moreover, inconsistent evaluation metrics and lenient answer matching criteria further obscure meaningful comparisons. In this work, we introduce a general method for constructing benchmarks, together with an evaluation protocol, to systematically assess KG-RAG methods under knowledge incompleteness. Our empirical results show that current KG-RAG methods have limited reasoning ability under missing knowledge, often rely on internal memorization, and exhibit varying degrees of generalization depending on their design.
A Hybrid Fully Convolutional CNN-Transformer Model for Inherently Interpretable Disease Detection from Retinal Fundus Images
In many medical imaging tasks, convolutional neural networks (CNNs) efficiently extract local features hierarchically. More recently, vision transformers (ViTs) have gained popularity, using self-attention mechanisms to capture global dependencies, but lacking the inherent spatial localization of convolutions. Therefore, hybrid models combining CNNs and ViTs have been developed to combine the strengths of both architectures. However, such hybrid models are difficult to interpret, which hinders their application in medical imaging. In this work, we introduce an interpretable-by-design hybrid fully convolutional CNN-Transformer architecture for retinal disease detection. Unlike widely used post-hoc saliency methods for ViTs, our approach generates faithful and localized evidence maps that directly reflect the mode's decision process. We evaluated our method on two medical tasks focused on disease detection using color fundus images. Our model achieves state-of-the-art predictive performance compared to black-box and interpretable models and provides class-specific sparse evidence maps in a single forward pass. The code is available at: https://github.com/kdjoumessi/Self-Explainable-CNN-Transformer.
comment: I made an error in this version
Documenting Deployment with Fabric: A Repository of Real-World AI Governance
Artificial intelligence (AI) is increasingly integrated into society, from financial services and traffic management to creative writing. Academic literature on the deployment of AI has mostly focused on the risks and harms that result from the use of AI. We introduce Fabric, a publicly available repository of deployed AI use cases to outline their governance mechanisms. Through semi-structured interviews with practitioners, we collect an initial set of 20 AI use cases. In addition, we co-design diagrams of the AI workflow with the practitioners. We discuss the oversight mechanisms and guardrails used in practice to safeguard AI use. The Fabric repository includes visual diagrams of AI use cases and descriptions of the deployed systems. Using the repository, we surface gaps in governance and find common patterns in human oversight of deployed AI systems. We intend for Fabric to serve as an extendable, evolving tool for researchers to study the effectiveness of AI governance.
comment: AIES 2025
Evaluating Knowledge Graph Based Retrieval Augmented Generation Methods under Knowledge Incompleteness
Knowledge Graph based Retrieval-Augmented Generation (KG-RAG) is a technique that enhances Large Language Model (LLM) inference in tasks like Question Answering (QA) by retrieving relevant information from knowledge graphs (KGs). However, real-world KGs are often incomplete, meaning that essential information for answering questions may be missing. Existing benchmarks do not adequately capture the impact of KG incompleteness on KG-RAG performance. In this paper, we systematically evaluate KG-RAG methods under incomplete KGs by removing triples using different methods and analyzing the resulting effects. We demonstrate that KG-RAG methods are sensitive to KG incompleteness, highlighting the need for more robust approaches in realistic settings.
comment: IRISAI'25
Dually Hierarchical Drift Adaptation for Online Configuration Performance Learning ICSE 2026
Modern configurable software systems need to learn models that correlate configuration and performance. However, when the system operates in dynamic environments, the workload variations, hardware changes, and system updates will inevitably introduce concept drifts at different levels - global drifts, which reshape the performance landscape of the entire configuration space; and local drifts, which only affect certain sub-regions of that space. As such, existing offline and transfer learning approaches can struggle to adapt to these implicit and unpredictable changes in real-time, rendering configuration performance learning challenging. To address this, we propose DHDA, an online configuration performance learning framework designed to capture and adapt to these drifts at different levels. The key idea is that DHDA adapts to both the local and global drifts using dually hierarchical adaptation: at the upper level, we redivide the data into different divisions, within each of which the local model is retrained, to handle global drifts only when necessary. At the lower level, the local models of the divisions can detect local drifts and adapt themselves asynchronously. To balance responsiveness and efficiency, DHDA combines incremental updates with periodic full retraining to minimize redundant computation when no drifts are detected. Through evaluating eight software systems and against state-of-the-art approaches, we show that DHDA achieves considerably better accuracy and can effectively adapt to drifts with up to 2x improvements, while incurring reasonable overhead and is able to improve different local models in handling concept drift.
comment: Accepted by ICSE 2026
Quantifying Fairness in LLMs Beyond Tokens: A Semantic and Statistical Perspective
Large Language Models (LLMs) often generate responses with inherent biases, undermining their reliability in real-world applications. Existing evaluation methods often overlook biases in long-form responses and the intrinsic variability of LLM outputs. To address these challenges, we propose FiSCo (Fine-grained Semantic Comparison), a novel statistical framework to evaluate group-level fairness in LLMs by detecting subtle semantic differences in long-form responses across demographic groups. Unlike prior work focusing on sentiment or token-level comparisons, FiSCo goes beyond surface-level analysis by operating at the claim level, leveraging entailment checks to assess the consistency of meaning across responses. We decompose model outputs into semantically distinct claims and apply statistical hypothesis testing to compare inter- and intra-group similarities, enabling robust detection of subtle biases. We formalize a new group counterfactual fairness definition and validate FiSCo on both synthetic and human-annotated datasets spanning gender, race, and age. Experiments show that FiSCo more reliably identifies nuanced biases while reducing the impact of stochastic LLM variability, outperforming various evaluation metrics.
comment: 29 pages, 9 figures, 15 tables
Quantized Neural Networks for Microcontrollers: A Comprehensive Review of Methods, Platforms, and Applications
The deployment of Quantized Neural Networks (QNNs) on resource-constrained devices, such as microcontrollers, has introduced significant challenges in balancing model performance, computational complexity, and memory constraints. Tiny Machine Learning (TinyML) addresses these issues by integrating advancements across machine learning algorithms, hardware acceleration, and software optimization to efficiently run deep neural networks on embedded systems. This survey presents a hardware-centric introduction to quantization, systematically reviewing essential quantization techniques employed to accelerate deep learning models for embedded applications. In particular, further emphasis is placed on the critical trade-offs between model performance and hardware capabilities. The survey further evaluates existing software frameworks and hardware platforms designed specifically for supporting QNN execution on microcontrollers. Moreover, we provide an analysis of the current challenges and an outline of promising future directions in the rapidly evolving domain of QNN deployment.
comment: 39 pages, 16 figures, 8 Tables, submitted to the Proceedings of the IEEE
WebInject: Prompt Injection Attack to Web Agents EMNLP 2025
Multi-modal large language model (MLLM)-based web agents interact with webpage environments by generating actions based on screenshots of the webpages. In this work, we propose WebInject, a prompt injection attack that manipulates the webpage environment to induce a web agent to perform an attacker-specified action. Our attack adds a perturbation to the raw pixel values of the rendered webpage. After these perturbed pixels are mapped into a screenshot, the perturbation induces the web agent to perform the attacker-specified action. We formulate the task of finding the perturbation as an optimization problem. A key challenge in solving this problem is that the mapping between raw pixel values and screenshot is non-differentiable, making it difficult to backpropagate gradients to the perturbation. To overcome this, we train a neural network to approximate the mapping and apply projected gradient descent to solve the reformulated optimization problem. Extensive evaluation on multiple datasets shows that WebInject is highly effective and significantly outperforms baselines.
comment: EMNLP 2025 main
Scientifically-Interpretable Reasoning Network (ScIReN): Discovering Hidden Relationships in the Carbon Cycle and Beyond
Understanding how carbon flows through the soil is crucial for mitigating the effects of climate change. While soils have potential to sequester carbon from the atmosphere, the soil carbon cycle remains poorly understood. Scientists have developed mathematical process-based models of the soil carbon cycle based on existing knowledge, but they contain numerous unknown parameters that must be set in an ad-hoc manner, and often fit observations poorly. On the other hand, neural networks can learn patterns from data, but do not respect known scientific laws, nor can they reveal novel scientific relationships due to their black-box nature. We thus propose Scientifically-Interpretable Reasoning Network (ScIReN), a fully-transparent framework that combines interpretable neural and process-based reasoning. An interpretable encoder predicts scientifically-meaningful latent parameters, which are then passed through a differentiable process-based decoder to predict labeled output variables. ScIReN leverages Kolmogorov-Arnold networks (KAN) to ensure the encoder is fully interpretable and reveals relationships between input features and latent parameters; it uses novel smoothness penalties to balance expressivity and simplicity. ScIReN also uses a novel hard-sigmoid constraint layer to restrict latent parameters to meaningful ranges defined by scientific prior knowledge. While the process-based decoder enforces established scientific knowledge, the KAN-based encoder reveals new scientific relationships hidden in conventional black-box models. We apply ScIReN on two tasks: simulating the flow of organic carbon through soils, and modeling ecosystem respiration from plants. In both tasks, ScIReN outperforms black-box networks in predictive accuracy while providing substantial scientific interpretability -- it can infer latent scientific mechanisms and their relationships with input features.
comment: 18 pages, 9 figures
Interpretable Mnemonic Generation for Kanji Learning via Expectation-Maximization EMNLP 2025
Learning Japanese vocabulary is a challenge for learners from Roman alphabet backgrounds due to script differences. Japanese combines syllabaries like hiragana with kanji, which are logographic characters of Chinese origin. Kanji are also complicated due to their complexity and volume. Keyword mnemonics are a common strategy to aid memorization, often using the compositional structure of kanji to form vivid associations. Despite recent efforts to use large language models (LLMs) to assist learners, existing methods for LLM-based keyword mnemonic generation function as a black box, offering limited interpretability. We propose a generative framework that explicitly models the mnemonic construction process as driven by a set of common rules, and learn them using a novel Expectation-Maximization-type algorithm. Trained on learner-authored mnemonics from an online platform, our method learns latent structures and compositional rules, enabling interpretable and systematic mnemonics generation. Experiments show that our method performs well in the cold-start setting for new learners while providing insight into the mechanisms behind effective mnemonic creation.
comment: The Conference on Empirical Methods in Natural Language Processing (EMNLP 2025)
Compression versus Accuracy: A Hierarchy of Lifted Models
Probabilistic graphical models that encode indistinguishable objects and relations among them use first-order logic constructs to compress a propositional factorised model for more efficient (lifted) inference. To obtain a lifted representation, the state-of-the-art algorithm Advanced Colour Passing (ACP) groups factors that represent matching distributions. In an approximate version using $\varepsilon$ as a hyperparameter, factors are grouped that differ by a factor of at most $(1\pm \varepsilon)$. However, finding a suitable $\varepsilon$ is not obvious and may need a lot of exploration, possibly requiring many ACP runs with different $\varepsilon$ values. Additionally, varying $\varepsilon$ can yield wildly different models, leading to decreased interpretability. Therefore, this paper presents a hierarchical approach to lifted model construction that is hyperparameter-free. It efficiently computes a hierarchy of $\varepsilon$ values that ensures a hierarchy of models, meaning that once factors are grouped together given some $\varepsilon$, these factors will be grouped together for larger $\varepsilon$ as well. The hierarchy of $\varepsilon$ values also leads to a hierarchy of error bounds. This allows for explicitly weighing compression versus accuracy when choosing specific $\varepsilon$ values to run ACP with and enables interpretability between the different models.
BudgetThinker: Empowering Budget-aware LLM Reasoning with Control Tokens
Recent advancements in Large Language Models (LLMs) have leveraged increased test-time computation to enhance reasoning capabilities, a strategy that, while effective, incurs significant latency and resource costs, limiting their applicability in real-world time-constrained or cost-sensitive scenarios. This paper introduces BudgetThinker, a novel framework designed to empower LLMs with budget-aware reasoning, enabling precise control over the length of their thought processes. We propose a methodology that periodically inserts special control tokens during inference to continuously inform the model of its remaining token budget. This approach is coupled with a comprehensive two-stage training pipeline, beginning with Supervised Fine-Tuning (SFT) to familiarize the model with budget constraints, followed by a curriculum-based Reinforcement Learning (RL) phase that utilizes a length-aware reward function to optimize for both accuracy and budget adherence. We demonstrate that BudgetThinker significantly surpasses strong baselines in maintaining performance across a variety of reasoning budgets on challenging mathematical benchmarks. Our method provides a scalable and effective solution for developing efficient and controllable LLM reasoning, making advanced models more practical for deployment in resource-constrained and real-time environments.
Time-RA: Towards Time Series Reasoning for Anomaly with LLM Feedback
Time series anomaly detection is critical across various domains, yet current approaches often limit analysis to mere binary anomaly classification without detailed categorization or further explanatory reasoning. To address these limitations, we propose a novel task, Time-series Reasoning for Anomaly (Time-RA) that transforms classical time series anomaly detection from a discriminative into a generative, reasoning-intensive task leveraging Large Language Models (LLMs). Also, we introduce the first real-world multimodal benchmark dataset, RATs40K, explicitly annotated for anomaly reasoning, comprising approximately 40,000 samples across 10 real-world domains. Each sample includes numeric time series data, contextual text information, and visual representations, each annotated with fine-grained categories (14 types for univariate anomalies and 6 for multivariate anomalies) and structured explanatory reasoning. We develop a sophisticated annotation framework utilizing ensemble-generated labels refined through GPT-4-driven feedback, ensuring accuracy and interpretability. Extensive benchmarking of LLMs and multimodal LLMs demonstrates the capabilities and limitations of current models, highlighting the critical role of supervised fine-tuning. Our dataset and task pave the way for significant advancements in interpretable time series anomaly detection and reasoning. The code (https://github.com/yyysjz1997/Time-RA) and dataset (https://huggingface.co/datasets/Time-RA/RATs40K) have been fully open-sourced to support and accelerate future research in this area.
comment: Under review. 19 pages, 8 figures, 12 tables. Code and dataset are publicly available
Alice's Adventures in a Differentiable Wonderland -- Volume I, A Tour of the Land
Neural networks surround us, in the form of large language models, speech transcription systems, molecular discovery algorithms, robotics, and much more. Stripped of anything else, neural networks are compositions of differentiable primitives, and studying them means learning how to program and how to interact with these models, a particular example of what is called differentiable programming. This primer is an introduction to this fascinating field imagined for someone, like Alice, who has just ventured into this strange differentiable wonderland. I overview the basics of optimizing a function via automatic differentiation, and a selection of the most common designs for handling sequences, graphs, texts, and audios. The focus is on a intuitive, self-contained introduction to the most important design techniques, including convolutional, attentional, and recurrent blocks, hoping to bridge the gap between theory and code (PyTorch and JAX) and leaving the reader capable of understanding some of the most advanced models out there, such as large language models (LLMs) and multimodal architectures.
comment: Companion website for additional chapters: https://www.sscardapane.it/alice-book
CMPhysBench: A Benchmark for Evaluating Large Language Models in Condensed Matter Physics
We introduce CMPhysBench, designed to assess the proficiency of Large Language Models (LLMs) in Condensed Matter Physics, as a novel Benchmark. CMPhysBench is composed of more than 520 graduate-level meticulously curated questions covering both representative subfields and foundational theoretical frameworks of condensed matter physics, such as magnetism, superconductivity, strongly correlated systems, etc. To ensure a deep understanding of the problem-solving process,we focus exclusively on calculation problems, requiring LLMs to independently generate comprehensive solutions. Meanwhile, leveraging tree-based representations of expressions, we introduce the Scalable Expression Edit Distance (SEED) score, which provides fine-grained (non-binary) partial credit and yields a more accurate assessment of similarity between prediction and ground-truth. Our results show that even the best models, Grok-4, reach only 36 average SEED score and 28% accuracy on CMPhysBench, underscoring a significant capability gap, especially for this practical and frontier domain relative to traditional physics. The code anddataset are publicly available at https://github.com/CMPhysBench/CMPhysBench.
comment: 29 pages, 7 figures
Robustness is Important: Limitations of LLMs for Data Fitting
Large Language Models (LLMs) are being applied in a wide array of settings, well beyond the typical language-oriented use cases. In particular, LLMs are increasingly used as a plug-and-play method for fitting data and generating predictions. Prior work has shown that LLMs, via in-context learning or supervised fine-tuning, can perform competitively with many tabular supervised learning techniques in terms of predictive performance. However, we identify a critical vulnerability of using LLMs for data fitting -- making changes to data representation that are completely irrelevant to the underlying learning task can drastically alter LLMs' predictions on the same data. For example, simply changing variable names can sway the size of prediction error by as much as 82% in certain settings. Such prediction sensitivity with respect to task-irrelevant variations manifests under both in-context learning and supervised fine-tuning, for both close-weight and open-weight general-purpose LLMs. Moreover, by examining the attention scores of an open-weight LLM, we discover a non-uniform attention pattern: training examples and variable names/values which happen to occupy certain positions in the prompt receive more attention when output tokens are generated, even though different positions are expected to receive roughly the same attention. This partially explains the sensitivity in the presence of task-irrelevant variations. We also consider a state-of-the-art tabular foundation model (TabPFN) trained specifically for data fitting. Despite being explicitly designed to achieve prediction robustness, TabPFN is still not immune to task-irrelevant variations. Overall, despite LLMs' impressive predictive capabilities, currently they lack even the basic level of robustness to be used as a principled data-fitting tool.
Guiding a diffusion model using sliding windows BMVC 2025
Guidance is a widely used technique for diffusion models to enhance sample quality. Technically, guidance is realised by using an auxiliary model that generalises more broadly than the primary model. Using a 2D toy example, we first show that it is highly beneficial when the auxiliary model exhibits similar but stronger generalisation errors than the primary model. Based on this insight, we introduce \emph{masked sliding window guidance (M-SWG)}, a novel, training-free method. M-SWG upweights long-range spatial dependencies by guiding the primary model with itself by selectively restricting its receptive field. M-SWG requires neither access to model weights from previous iterations, additional training, nor class conditioning. M-SWG achieves a superior Inception score (IS) compared to previous state-of-the-art training-free approaches, without introducing sample oversaturation. In conjunction with existing guidance methods, M-SWG reaches state-of-the-art Frechet DINOv2 distance on ImageNet using EDM2-XXL and DiT-XL. The code is available at https://github.com/HHU-MMBS/swg_bmvc2025_official.
comment: Accepted at BMVC 2025. 30 pages, 16 figures in total, including appendix
Invited Paper: Feature-to-Classifier Co-Design for Mixed-Signal Smart Flexible Wearables for Healthcare at the Extreme Edge
Flexible Electronics (FE) offer a promising alternative to rigid silicon-based hardware for wearable healthcare devices, enabling lightweight, conformable, and low-cost systems. However, their limited integration density and large feature sizes impose strict area and power constraints, making ML-based healthcare systems-integrating analog frontend, feature extraction and classifier-particularly challenging. Existing FE solutions often neglect potential system-wide solutions and focus on the classifier, overlooking the substantial hardware cost of feature extraction and Analog-to-Digital Converters (ADCs)-both major contributors to area and power consumption. In this work, we present a holistic mixed-signal feature-to-classifier co-design framework for flexible smart wearable systems. To the best of our knowledge, we design the first analog feature extractors in FE, significantly reducing feature extraction cost. We further propose an hardware-aware NAS-inspired feature selection strategy within ML training, enabling efficient, application-specific designs. Our evaluation on healthcare benchmarks shows our approach delivers highly accurate, ultra-area-efficient flexible systems-ideal for disposable, low-power wearable monitoring.
comment: Accepted for publication at 2025 IEEE/ACM International Conference on Computer-Aided Design (ICCAD), Oct. 26-30 2025, Munich, DE
Arbitrary Precision Printed Ternary Neural Networks with Holistic Evolutionary Approximation
Printed electronics offer a promising alternative for applications beyond silicon-based systems, requiring properties like flexibility, stretchability, conformality, and ultra-low fabrication costs. Despite the large feature sizes in printed electronics, printed neural networks have attracted attention for meeting target application requirements, though realizing complex circuits remains challenging. This work bridges the gap between classification accuracy and area efficiency in printed neural networks, covering the entire processing-near-sensor system design and co-optimization from the analog-to-digital interface-a major area and power bottleneck-to the digital classifier. We propose an automated framework for designing printed Ternary Neural Networks with arbitrary input precision, utilizing multi-objective optimization and holistic approximation. Our circuits outperform existing approximate printed neural networks by 17x in area and 59x in power on average, being the first to enable printed-battery-powered operation with under 5% accuracy loss while accounting for analog-to-digital interfacing costs.
comment: Accepted at IEEE Transactions on Circuits and Systems for Artificial Intelligence (TCASAI)
Beyond Frequency: The Role of Redundancy in Large Language Model Memorization
Memorization in large language models poses critical risks for privacy and fairness as these systems scale to billions of parameters. While previous studies established correlations between memorization and factors like token frequency and repetition patterns, we revealed distinct response patterns: frequency increases minimally impact memorized samples (e.g. 0.09) while substantially affecting non-memorized samples (e.g., 0.25), with consistency observed across model scales. Through counterfactual analysis by perturbing sample prefixes and quantifying perturbation strength through token positional changes, we demonstrate that redundancy correlates with memorization patterns. Our findings establish that: about 79% of memorized samples are low-redundancy, these low-redundancy samples exhibit 2-fold higher vulnerability than high-redundancy ones, and consequently memorized samples drop by 0.6 under perturbation while non-memorized samples drop by only 0.01, indicating that more redundant content becomes both more memorable and more fragile. These findings suggest potential redundancy-guided approaches for data preprocessing, thereby reducing privacy risks and mitigating bias to ensure fairness in model deployments.
comment: 8 figures
BASE-Q: Bias and Asymmetric Scaling Enhanced Rotational Quantization for Large Language Models
Rotations have become essential to state-of-the-art quantization pipelines for large language models (LLMs) by effectively smoothing outliers in weights and activations. However, further optimizing the rotation parameters offers only limited performance gains and introduces significant training overhead: due to rotation parameter sharing, full-model must be loaded simultaneously to enable backpropagation, resulting in substantial memory consumption and limited practical utility. In this work, we identify two fundamental limitations of current rotational quantization methods: (i) rotation fails to align channel means, resulting in wider quantization bounds and increased rounding errors; and (ii) rotation makes the activation distribution more Gaussian-like, increasing energy loss caused by clipping errors. To address these issues, we introduce \textbf{BASE-Q}, a simple yet powerful approach that combines bias correction and asymmetric scaling to effectively reduce rounding and clipping errors. Furthermore, BASE-Q enables blockwise optimization, eliminating the need for memory-intensive full-model backpropagation. Extensive experiments on various LLMs and benchmarks demonstrate the effectiveness of BASE-Q, narrowing the accuracy gap to full-precision models by 50.5\%, 42.9\%, and 29.2\% compared to QuaRot, SpinQuant, and OSTQuant, respectively. The code will be released soon.
Categorical Data Clustering via Value Order Estimated Distance Metric Learning SIGMOD 2026
Clustering is a popular machine learning technique for data mining that can process and analyze datasets to automatically reveal sample distribution patterns. Since the ubiquitous categorical data naturally lack a well-defined metric space such as the Euclidean distance space of numerical data, the distribution of categorical data is usually under-represented, and thus valuable information can be easily twisted in clustering. This paper, therefore, introduces a novel order distance metric learning approach to intuitively represent categorical attribute values by learning their optimal order relationship and quantifying their distance in a line similar to that of the numerical attributes. Since subjectively created qualitative categorical values involve ambiguity and fuzziness, the order distance metric is learned in the context of clustering. Accordingly, a new joint learning paradigm is developed to alternatively perform clustering and order distance metric learning with low time complexity and a guarantee of convergence. Due to the clustering-friendly order learning mechanism and the homogeneous ordinal nature of the order distance and Euclidean distance, the proposed method achieves superior clustering accuracy on categorical and mixed datasets. More importantly, the learned order distance metric greatly reduces the difficulty of understanding and managing the non-intuitive categorical data. Experiments with ablation studies, significance tests, case studies, etc., have validated the efficacy of the proposed method. The source code is available at https://github.com/DAJ0612/OCL_Source_Code.
comment: Accepted to SIGMOD 2026
A Collaborative Content Moderation Framework for Toxicity Detection based on Conformalized Estimates of Annotation Disagreement
Content moderation typically combines the efforts of human moderators and machine learning models. However, these systems often rely on data where significant disagreement occurs during moderation, reflecting the subjective nature of toxicity perception. Rather than dismissing this disagreement as noise, we interpret it as a valuable signal that highlights the inherent ambiguity of the content,an insight missed when only the majority label is considered. In this work, we introduce a novel content moderation framework that emphasizes the importance of capturing annotation disagreement. Our approach uses multitask learning, where toxicity classification serves as the primary task and annotation disagreement is addressed as an auxiliary task. Additionally, we leverage uncertainty estimation techniques, specifically Conformal Prediction, to account for both the ambiguity in comment annotations and the model's inherent uncertainty in predicting toxicity and disagreement.The framework also allows moderators to adjust thresholds for annotation disagreement, offering flexibility in determining when ambiguity should trigger a review. We demonstrate that our joint approach enhances model performance, calibration, and uncertainty estimation, while offering greater parameter efficiency and improving the review process in comparison to single-task methods.
comment: 35 pages, 1 figure
DisCoPatch: Taming Adversarially-driven Batch Statistics for Improved Out-of-Distribution Detection ICCV 2025
Out-of-distribution (OOD) detection holds significant importance across many applications. While semantic and domain-shift OOD problems are well-studied, this work focuses on covariate shifts - subtle variations in the data distribution that can degrade machine learning performance. We hypothesize that detecting these subtle shifts can improve our understanding of in-distribution boundaries, ultimately improving OOD detection. In adversarial discriminators trained with Batch Normalization (BN), real and adversarial samples form distinct domains with unique batch statistics - a property we exploit for OOD detection. We introduce DisCoPatch, an unsupervised Adversarial Variational Autoencoder (VAE) framework that harnesses this mechanism. During inference, batches consist of patches from the same image, ensuring a consistent data distribution that allows the model to rely on batch statistics. DisCoPatch uses the VAE's suboptimal outputs (generated and reconstructed) as negative samples to train the discriminator, thereby improving its ability to delineate the boundary between in-distribution samples and covariate shifts. By tightening this boundary, DisCoPatch achieves state-of-the-art results in public OOD detection benchmarks. The proposed model not only excels in detecting covariate shifts, achieving 95.5% AUROC on ImageNet-1K(-C) but also outperforms all prior methods on public Near-OOD (95.0%) benchmarks. With a compact model size of 25MB, it achieves high OOD detection performance at notably lower latency than existing methods, making it an efficient and practical solution for real-world OOD detection applications. The code is publicly available.
comment: ICCV 2025
A Hybrid Artificial Intelligence Method for Estimating Flicker in Power Systems
This paper introduces a novel hybrid AI method combining H filtering and an adaptive linear neuron network for flicker component estimation in power distribution systems.The proposed method leverages the robustness of the H filter to extract the voltage envelope under uncertain and noisy conditions followed by the use of ADALINE to accurately identify flicker frequencies embedded in the envelope.This synergy enables efficient time domain estimation with rapid convergence and noise resilience addressing key limitations of existing frequency domain approaches.Unlike conventional techniques this hybrid AI model handles complex power disturbances without prior knowledge of noise characteristics or extensive training.To validate the method performance we conduct simulation studies based on IEC Standard 61000 4 15 supported by statistical analysis Monte Carlo simulations and real world data.Results demonstrate superior accuracy robustness and reduced computational load compared to Fast Fourier Transform and Discrete Wavelet Transform based estimators.
comment: 31 pages, 12 figures, and 6 tables
Toxicity Begets Toxicity: Unraveling Conversational Chains in Political Podcasts
Tackling toxic behavior in digital communication continues to be a pressing concern for both academics and industry professionals. While significant research has explored toxicity on platforms like social networks and discussion boards, podcasts despite their rapid rise in popularity remain relatively understudied in this context. This work seeks to fill that gap by curating a dataset of political podcast transcripts and analyzing them with a focus on conversational structure. Specifically, we investigate how toxicity surfaces and intensifies through sequences of replies within these dialogues, shedding light on the organic patterns by which harmful language can escalate across conversational turns. Warning: Contains potentially abusive/toxic contents.
Atom-Searcher: Enhancing Agentic Deep Research via Fine-Grained Atomic Thought Reward
Large language models (LLMs) exhibit remarkable problem-solving abilities, but struggle with complex tasks due to static internal knowledge. Retrieval-Augmented Generation (RAG) enhances access to external information, yet remains limited in multi-hop reasoning and strategic search due to rigid workflows. Recent advancements in agentic deep research empower LLMs to autonomously reason, search, and synthesize information. However, current approaches relying on outcome-based reinforcement learning (RL) face critical issues such as conflicting gradients and reward sparsity, limiting performance gains and training efficiency. To address these, we first propose Atomic Thought, a novel LLM thinking paradigm that decomposes reasoning into fine-grained functional units. These units are supervised by Reasoning Reward Models (RRMs), which provide Atomic Thought Rewards (ATR) for fine-grained guidance. Building on this, we propose Atom-Searcher, a novel RL framework for agentic deep research that integrates Atomic Thought and ATR. Atom-Searcher uses a curriculum-inspired reward schedule, prioritizing process-level ATR early and transitioning to outcome rewards, accelerating convergence on effective reasoning paths. Experiments on seven benchmarks show consistent improvements over the state-of-the-art. Key advantages include: (1) Atom-Searcher scales computation at test-time. (2) Atomic Thought provides supervision anchors for RRMs, bridging deep research tasks and RRMs. (3) Atom-Searcher exhibits more interpretable, human-like reasoning patterns.
DeepTrans: Deep Reasoning Translation via Reinforcement Learning ACL
Recently, deep reasoning LLMs (e.g., OpenAI o1 and DeepSeek-R1) have shown promising performance in various downstream tasks. Free translation is an important and interesting task in the multilingual world, which requires going beyond word-for-word translation. However, the task is still under-explored in deep reasoning LLMs. In this paper, we introduce DeepTrans, a deep reasoning translation model that learns free translation via reinforcement learning (RL). Specifically, we carefully build a reward model with pre-defined scoring criteria on both the translation results and the thought processes. The reward model teaches DeepTrans how to think and free-translate the given sentences during RL. Besides, our RL training does not need any labeled translations, avoiding the human-intensive annotation or resource-intensive data synthesis. Experimental results show the effectiveness of DeepTrans. Using Qwen2.5-7B as the backbone, DeepTrans improves performance by 16.3% in literature translation, and outperforms strong deep reasoning LLMs. Moreover, we summarize the failures and interesting findings during our RL exploration. We hope this work could inspire other researchers in free translation.
comment: Accepted by Transactions of the Association for Computational Linguistics (TACL)
Retrieval-Augmented Machine Translation with Unstructured Knowledge EMNLP 2025
Retrieval-augmented generation (RAG) introduces additional information to enhance large language models (LLMs). In machine translation (MT), previous work typically retrieves in-context examples from paired MT corpora, or domain-specific knowledge from knowledge graphs, to enhance MT models. However, a large amount of world knowledge is organized in unstructured documents, and might not be fully paired across different languages. In this paper, we study retrieval-augmented MT using unstructured documents. Specifically, we build RAGtrans, the first benchmark to train and evaluate LLMs' retrieval-augmented MT ability. RAGtrans contains 169K MT samples collected via GPT-4o and human translators. Besides, documents from various languages are also provided to supply the knowledge to these samples. Based on RAGtrans, we further propose a multi-task training method to teach LLMs how to use information from multilingual documents during their translation. The method uses existing multilingual corpora to create auxiliary training objectives without additional labeling requirements. Extensive experiments show that the method improves LLMs by 1.6-3.1 BLEU and 1.0-2.0 COMET scores in En-Zh, and 1.7-2.9 BLEU and 2.1-2.7 COMET scores in En-De. We also conclude the critical difficulties that current LLMs face with this task.
comment: EMNLP 2025 Findings
Towards Understanding Camera Motions in Any Video
We introduce CameraBench, a large-scale dataset and benchmark designed to assess and improve camera motion understanding. CameraBench consists of ~3,000 diverse internet videos, annotated by experts through a rigorous multi-stage quality control process. One of our contributions is a taxonomy of camera motion primitives, designed in collaboration with cinematographers. We find, for example, that some motions like "follow" (or tracking) require understanding scene content like moving subjects. We conduct a large-scale human study to quantify human annotation performance, revealing that domain expertise and tutorial-based training can significantly enhance accuracy. For example, a novice may confuse zoom-in (a change of intrinsics) with translating forward (a change of extrinsics), but can be trained to differentiate the two. Using CameraBench, we evaluate Structure-from-Motion (SfM) and Video-Language Models (VLMs), finding that SfM models struggle to capture semantic primitives that depend on scene content, while VLMs struggle to capture geometric primitives that require precise estimation of trajectories. We then fine-tune a generative VLM on CameraBench to achieve the best of both worlds and showcase its applications, including motion-augmented captioning, video question answering, and video-text retrieval. We hope our taxonomy, benchmark, and tutorials will drive future efforts towards the ultimate goal of understanding camera motions in any video.
comment: Project site: https://linzhiqiu.github.io/papers/camerabench/
TrueGL: A Truthful, Reliable, and Unified Engine for Grounded Learning in Full-Stack Search
In the age of open and free information, a concerning trend of reliance on AI is emerging. However, existing AI tools struggle to evaluate the credibility of information and to justify their assessments. Hence, there is a growing need for systems that can help users evaluate the trustworthiness of online information. Although major search engines incorporate AI features, they often lack clear reliability indicators. We present TrueGL, a model that makes trustworthy search results more accessible. The model is a fine-tuned version of IBM's Granite-1B, trained on the custom dataset and integrated into a search engine with a reliability scoring system. We evaluate the system using prompt engineering and assigning each statement a continuous reliability score from 0.1 to 1, then instructing the model to return a textual explanation alongside the score. Each model's predicted scores are measured against real scores using standard evaluation metrics. TrueGL consistently outperforms other small-scale LLMs and rule-based approaches across all experiments on key evaluation metrics, including MAE, RMSE, and R2. The model's high accuracy, broad content coverage, and ease of use make trustworthy information more accessible and help reduce the spread of false or misleading content online. Our code is publicly available at https://github.com/AlgazinovAleksandr/TrueGL, and our model is publicly released at https://huggingface.co/JoydeepC/trueGL.
Unified Path Planner with Adaptive Safety and Optimality
Path planning for autonomous robots presents a fundamental trade-off between optimality and safety. While conventional algorithms typically prioritize one of these objectives, we introduce the Unified Path Planner (UPP), a unified framework that simultaneously addresses both. UPP is a graph-search-based algorithm that employs a modified heuristic function incorporating a dynamic safety cost, enabling an adaptive balance between path length and obstacle clearance. We establish theoretical sub-optimality bounds for the planner and demonstrate that its safety-to-optimality ratio can be tuned via adjustable parameters, with a trade-off in computational complexity. Extensive simulations show that UPP achieves a high success rate, generating near-optimal paths with only a negligible increase in cost over traditional A*, while ensuring safety margins that closely approach those of the classical Voronoi planner. Finally, the practical efficacy of UPP is validated through a hardware implementation on a TurtleBot, confirming its ability to navigate cluttered environments by generating safe, sub-optimal paths.
comment: 6 pages,4 figures
THEME: Enhancing Thematic Investing with Semantic Stock Representations and Temporal Dynamics CIKM
Thematic investing, which aims to construct portfolios aligned with structural trends, remains a challenging endeavor due to overlapping sector boundaries and evolving market dynamics. A promising direction is to build semantic representations of investment themes from textual data. However, despite their power, general-purpose LLM embedding models are not well-suited to capture the nuanced characteristics of financial assets, since the semantic representation of investment assets may differ fundamentally from that of general financial text. To address this, we introduce THEME, a framework that fine-tunes embeddings using hierarchical contrastive learning. THEME aligns themes and their constituent stocks using their hierarchical relationship, and subsequently refines these embeddings by incorporating stock returns. This process yields representations effective for retrieving thematically aligned assets with strong return potential. Empirical results demonstrate that THEME excels in two key areas. For thematic asset retrieval, it significantly outperforms leading large language models. Furthermore, its constructed portfolios demonstrate compelling performance. By jointly modeling thematic relationships from text and market dynamics from returns, THEME generates stock embeddings specifically tailored for a wide range of practical investment applications.
comment: Accepted at ACM International Conference on Information and Knowledge Management (CIKM)
TrustGeoGen: Formal-Verified Data Engine for Trustworthy Multi-modal Geometric Problem Solving
Mathematical geometric problem solving (GPS) demands verifiable logical coherence and multimodal reasoning capabilities. While large language models (LLMs) have shown rapid progress in GPS, their advancement is hindered by the lack of reliable benchmarks and systematic methodologies. A critical challenge is the inherent hallucination in LLMs, which leads to synthetic GPS datasets that are often noisy, unverified, and self-contradictory. To address this, we introduce TrustGeoGen, a data engine that generates formally verified geometric problems to establish a principled and trustworthy benchmark. Our engine integrates four key innovations: 1) Multimodal Alignment, which synchronizes the generation of diagrams, text, and step-by-step solutions; 2) Formal Verification, ensuring all reasoning paths are rule-compliant; 3) Connection Thinking, bridging formal deduction with human-like logical steps; and 4) our \textit{GeoExplore} series algorithms, which produce diverse problem variants with multiple solutions and self-reflective backtracking. Using this engine, we create the GeoTrust-200K dataset and the corresponding GeoTrust-test benchmark, both with guaranteed cross-modal integrity. Experiments reveal that state-of-the-art models achieve only 45.83\% accuracy on GeoTrust-test, highlighting its significant challenge. Furthermore, training on our synthesized data substantially improves model performance on GPS tasks, with strong generalization to out-of-domain (OOD) benchmarks. Our code and data are available at https://github.com/Alpha-Innovator/TrustGeoGen
Large Intestine 3D Shape Refinement Using Point Diffusion Models for Digital Phantom Generation
Accurate 3D modeling of human organs is critical for constructing digital phantoms in virtual imaging trials. However, organs such as the large intestine remain particularly challenging due to their complex geometry and shape variability. We propose CLAP, a novel Conditional LAtent Point-diffusion model that combines geometric deep learning with denoising diffusion models to enhance 3D representations of the large intestine. Given point clouds sampled from segmentation masks, we employ a hierarchical variational autoencoder to learn both global and local latent shape representations. Two conditional diffusion models operate within this latent space to refine the organ shape. A pretrained surface reconstruction model is then used to convert the refined point clouds into meshes. CLAP achieves substantial improvements in shape modeling accuracy, reducing Chamfer distance by 26% and Hausdorff distance by 36% relative to the initial suboptimal shapes. This approach offers a robust and extensible solution for high-fidelity organ modeling, with potential applicability to a wide range of anatomical structures.
ETTRL: Balancing Exploration and Exploitation in LLM Test-Time Reinforcement Learning Via Entropy Mechanism
Recent advancements in Large Language Models have yielded significant improvements in complex reasoning tasks such as mathematics and programming. However, these models remain heavily dependent on annotated data and exhibit limited adaptability in unsupervised scenarios. To address these limitations, test-time reinforcement learning (TTRL) has been proposed, which enables self-optimization by leveraging model-generated pseudo-labels. Despite its promise, TTRL faces several key challenges, including high inference costs due to parallel rollouts and early-stage estimation bias that fosters overconfidence, reducing output diversity and causing performance plateaus. To address these challenges, we introduce an entropy-based mechanism to enhance the exploration-exploitation balance in test-time reinforcement learning through two strategies: Entropy-fork Tree Majority Rollout (ETMR) and Entropy-based Advantage Reshaping (EAR). Compared with the baseline, our approach enables Llama3.1-8B to achieve a 68 percent relative improvement in Pass at 1 metric on the AIME 2024 benchmark, while consuming only 60 percent of the rollout tokens budget. This highlights our method's ability to effectively optimize the trade-off between inference efficiency, diversity, and estimation robustness, thereby advancing unsupervised reinforcement learning for open-domain reasoning tasks.
MAC-Tuning: LLM Multi-Compositional Problem Reasoning with Enhanced Knowledge Boundary Awareness EMNLP 2025
The hallucination of non-existent facts by LLMs is an important problem given its widespread adoption across various applications. Previous research addresses this problem by analyzing the internal parameterized knowledge boundaries to estimate confidence. However, these studies focus on the single-problem setting and have not explored the more challenging multi-problem setting, which requires accurately answering multiple questions simultaneously. We introduce a novel method for the multi-problem setting, Multiple Answers and Confidence Stepwise Tuning (MAC-Tuning), that separates the learning of answer prediction and confidence estimation during fine-tuning on instruction data. Extensive experiments demonstrate that our method outperforms baselines by up to 25\% in average precision.
comment: We release our code and resource at https://github.com/no-touch-fish/Multi-QA-Tuning. The paper is accepted into EMNLP 2025 main
FedSEA-LLaMA: A Secure, Efficient and Adaptive Federated Splitting Framework for Large Language Models
Private data holds promise for improving LLMs due to its high quality, but its scattered distribution across data silos and the high computational demands of LLMs limit their deployment in federated environments. To address this, the transformer-based federated split models are proposed, which offload most model parameters to the server (or distributed clients) while retaining only a small portion on the client to ensure data privacy. Despite this design, they still face three challenges: 1) Peer-to-peer key encryption struggles to secure transmitted vectors effectively; 2) The auto-regressive nature of LLMs means that federated split learning can only train and infer sequentially, causing high communication overhead; 3) Fixed partition points lack adaptability to downstream tasks. In this paper, we introduce FedSEA-LLaMA, a Secure, Efficient, and Adaptive Federated splitting framework based on LLaMA2. First, we inject Gaussian noise into forward-pass hidden states to enable secure end-to-end vector transmission. Second, we employ attention-mask compression and KV cache collaboration to reduce communication costs, accelerating training and inference. Third, we allow users to dynamically adjust the partition points for input/output blocks based on specific task requirements. Experiments on natural language understanding, summarization, and conversational QA tasks show that FedSEA-LLaMA maintains performance comparable to centralized LLaMA2 and achieves up to 8x speedups in training and inference. Further analysis of privacy attacks and different partition points also demonstrates the effectiveness of FedSEA-LLaMA in security and adaptability.
Latent Adaptive Planner for Dynamic Manipulation
We present the Latent Adaptive Planner (LAP), a trajectory-level latent-variable policy for dynamic nonprehensile manipulation (e.g., box catching) that formulates planning as inference in a low-dimensional latent space and is learned effectively from human demonstration videos. During execution, LAP achieves real-time adaptation by maintaining a posterior over the latent plan and performing variational replanning as new observations arrive. To bridge the embodiment gap between humans and robots, we introduce a model-based proportional mapping that regenerates accurate kinematic-dynamic joint states and object positions from human demonstrations. Through challenging box catching experiments with varying object properties, LAP demonstrates superior success rates, trajectory smoothness, and energy efficiency by learning human-like compliant motions and adaptive behaviors. Overall, LAP enables dynamic manipulation with real-time adaptation and successfully transfer across heterogeneous robot platforms using the same human demonstration videos.
DDaTR: Dynamic Difference-aware Temporal Residual Network for Longitudinal Radiology Report Generation
Radiology Report Generation (RRG) automates the creation of radiology reports from medical imaging, enhancing the efficiency of the reporting process. Longitudinal Radiology Report Generation (LRRG) extends RRG by incorporating the ability to compare current and prior exams, facilitating the tracking of temporal changes in clinical findings. Existing LRRG approaches only extract features from prior and current images using a visual pre-trained encoder, which are then concatenated to generate the final report. However, these methods struggle to effectively capture both spatial and temporal correlations during the feature extraction process. Consequently, the extracted features inadequately capture the information of difference across exams and thus underrepresent the expected progressions, leading to sub-optimal performance in LRRG. To address this, we develop a novel dynamic difference-aware temporal residual network (DDaTR). In DDaTR, we introduce two modules at each stage of the visual encoder to capture multi-level spatial correlations. The Dynamic Feature Alignment Module (DFAM) is designed to align prior features across modalities for the integrity of prior clinical information. Prompted by the enriched prior features, the dynamic difference-aware module (DDAM) captures favorable difference information by identifying relationships across exams. Furthermore, our DDaTR employs the dynamic residual network to unidirectionally transmit longitudinal information, effectively modelling temporal correlations. Extensive experiments demonstrated superior performance over existing methods on three benchmarks, proving its efficacy in both RRG and LRRG tasks.
comment: Accepted in IEEE Transactions on Medical Imaging (TMI). Code is available at https://github.com/xmed-lab/DDaTR
QHackBench: Benchmarking Large Language Models for Quantum Code Generation Using PennyLane Hackathon Challenges
Recent advances in Large Language Models (LLMs) have demonstrated strong potential in code generation, yet their effectiveness in quantum computing remains underexplored. This paper benchmarks LLMs for PennyLane-based quantum code generation using real-world challenges from the Quantum Hackathon (QHack). We introduce QHackBench, a novel benchmark dataset derived from QHack competitions, and evaluate model performance under vanilla prompting and Retrieval-Augmented Generation (RAG). Our structured evaluation framework assesses functional correctness, syntactic validity, and execution success across varying challenge difficulties. Results indicate that RAG-enhanced models, supplemented with an augmented PennyLane dataset, approximately generate similar results as the standard prompting, particularly in complex quantum algorithms. Additionally, we introduce a multi-agent evaluation pipeline that iteratively refines incorrect solutions, further enhancing execution success rates. To foster further research, we commit to publicly releasing QHackBench, along with our evaluation framework and experimental results, enabling continued advancements in AI-assisted quantum programming.
comment: To appear at the IEEE International Conference on Quantum Artificial Intelligence (QAI), Naples, Italy, November 2025
SKA-Bench: A Fine-Grained Benchmark for Evaluating Structured Knowledge Understanding of LLMs EMNLP 2025
Although large language models (LLMs) have made significant progress in understanding Structured Knowledge (SK) like KG and Table, existing evaluations for SK understanding are non-rigorous (i.e., lacking evaluations of specific capabilities) and focus on a single type of SK. Therefore, we aim to propose a more comprehensive and rigorous structured knowledge understanding benchmark to diagnose the shortcomings of LLMs. In this paper, we introduce SKA-Bench, a Structured Knowledge Augmented QA Benchmark that encompasses four widely used structured knowledge forms: KG, Table, KG+Text, and Table+Text. We utilize a three-stage pipeline to construct SKA-Bench instances, which includes a question, an answer, positive knowledge units, and noisy knowledge units. To evaluate the SK understanding capabilities of LLMs in a fine-grained manner, we expand the instances into four fundamental ability testbeds: Noise Robustness, Order Insensitivity, Information Integration, and Negative Rejection. Empirical evaluations on 8 representative LLMs, including the advanced DeepSeek-R1, indicate that existing LLMs still face significant challenges in understanding structured knowledge, and their performance is influenced by factors such as the amount of noise, the order of knowledge units, and hallucination phenomenon. Our dataset and code are available at https://github.com/zjukg/SKA-Bench.
comment: EMNLP 2025
Trust but Verify! A Survey on Verification Design for Test-time Scaling
Test-time scaling (TTS) has emerged as a new frontier for scaling the performance of Large Language Models. In test-time scaling, by using more computational resources during inference, LLMs can improve their reasoning process and task performance. Several approaches have emerged for TTS such as distilling reasoning traces from another model or exploring the vast decoding search space by employing a verifier. The verifiers serve as reward models that help score the candidate outputs from the decoding process to diligently explore the vast solution space and select the best outcome. This paradigm commonly termed has emerged as a superior approach owing to parameter free scaling at inference time and high performance gains. The verifiers could be prompt-based, fine-tuned as a discriminative or generative model to verify process paths, outcomes or both. Despite their widespread adoption, there is no detailed collection, clear categorization and discussion of diverse verification approaches and their training mechanisms. In this survey, we cover the diverse approaches in the literature and present a unified view of verifier training, types and their utility in test-time scaling. Our repository can be found at https://github.com/elixir-research-group/Verifierstesttimescaling.github.io.
comment: 18 pages
Adaptive Duration Model for Text Speech Alignment
Speech-to-text alignment is a critical component of neural text to speech (TTS) models. Autoregressive TTS models typically use an attention mechanism to learn these alignments on-line, while non-autoregressive end to end TTS models rely on durations extracted from external sources. In this paper, we propose a novel duration prediction framework that can give promising phoneme-level duration distribution with given text. In our experiments, the proposed duration model has more precise prediction and adaptation ability to conditions, compared to previous baseline models. Specifically, it makes a considerable improvement on phoneme-level alignment accuracy and makes the performance of zero-shot TTS models more robust to the mismatch between prompt audio and input audio.
comment: 4 pages
RevPRAG: Revealing Poisoning Attacks in Retrieval-Augmented Generation through LLM Activation Analysis EMNLP 2025
Retrieval-Augmented Generation (RAG) enriches the input to LLMs by retrieving information from the relevant knowledge database, enabling them to produce responses that are more accurate and contextually appropriate. It is worth noting that the knowledge database, being sourced from publicly available channels such as Wikipedia, inevitably introduces a new attack surface. RAG poisoning involves injecting malicious texts into the knowledge database, ultimately leading to the generation of the attacker's target response (also called poisoned response). However, there are currently limited methods available for detecting such poisoning attacks. We aim to bridge the gap in this work. Particularly, we introduce RevPRAG, a flexible and automated detection pipeline that leverages the activations of LLMs for poisoned response detection. Our investigation uncovers distinct patterns in LLMs' activations when generating correct responses versus poisoned responses. Our results on multiple benchmark datasets and RAG architectures show our approach could achieve 98% true positive rate, while maintaining false positive rates close to 1%.
comment: Accepted to Findings of EMNLP 2025
Mamba State-Space Models Are Lyapunov-Stable Learners
Mamba state-space models (SSMs) have recently outperformed state-of-the-art (SOTA) Transformer large language models (LLMs) in various tasks and been widely adapted. However, a major concern for stable learning in recurrent-based deep models (such as SSMs) is the sensitivity of their recurrent dynamics. Despite widespread adaptation, the sensitivity of Mamba's recurrent dynamics under common fine-tuning methods-e.g., mixed-precision fine-tuning (MPFT) and parameter-efficient fine-tuning (PEFT)-remains unexplored. Empirically, we show that Mamba LLMs are extremely stable to changes introduced by combinations of MPFT and PEFT, in stark contrast to Transformer LLMs, which we demonstrate may drastically diverge from their respective full-precision counterparts under different combinations of MPFT and PEFT (despite the near-ubiquitous adaptation of these fine-tuning frameworks for attention-based models). The demonstrated robustness of Mamba LLMs are due to their recurrent dynamics, which we prove are guaranteed to be stable using dynamical systems theory (in particular, Lyapunov stability). We conclude by using MPFT and PEFT to novelly study Mamba LLMs' in-context learning (ICL) abilities on natural language tasks, thus supplementing other recent work.
comment: TMLR, 27 pages, 12 figures, 4 tables
Region-Level Context-Aware Multimodal Understanding
Despite significant progress, existing research on Multimodal Large Language Models (MLLMs) mainly focuses on general visual understanding, overlooking the ability to integrate textual context associated with objects for a more context-aware multimodal understanding -- an ability we refer to as Region-level Context-aware Multimodal Understanding (RCMU). To address this limitation, we first formulate the RCMU task, which requires models to respond to user instructions by integrating both image content and textual information of regions or objects. To equip MLLMs with RCMU capabilities, we propose Region-level Context-aware Visual Instruction Tuning (RCVIT), which incorporates object information into the model input and enables the model to utilize bounding box coordinates to effectively associate objects' visual content with their textual information. To address the lack of datasets, we introduce the RCMU dataset, a large-scale visual instruction tuning dataset that covers multiple RCMU tasks. We also propose RC\&P-Bench, a comprehensive benchmark that can evaluate the performance of MLLMs in RCMU and multimodal personalized understanding tasks. Additionally, we propose a reference-free evaluation metric to perform a comprehensive and fine-grained evaluation of the region-level context-aware image descriptions. By performing RCVIT on Qwen2-VL models with the RCMU dataset, we developed RC-Qwen2-VL models. Experimental results indicate that RC-Qwen2-VL models not only achieve outstanding performance on multiple RCMU tasks but also demonstrate successful applications in multimodal RAG and personalized conversation. Our data, model and benchmark are available at https://github.com/hongliang-wei/RC-MLLM
comment: 12 pages, 6 figures
CE-RS-SBCIT A Novel Channel Enhanced Hybrid CNN Transformer with Residual, Spatial, and Boundary-Aware Learning for Brain Tumor MRI Analysis
Brain tumors remain among the most lethal human diseases, where early detection and accurate classification are critical for effective diagnosis and treatment planning. Although deep learning-based computer-aided diagnostic (CADx) systems have shown remarkable progress. However, conventional convolutional neural networks (CNNs) and Transformers face persistent challenges, including high computational cost, sensitivity to minor contrast variations, structural heterogeneity, and texture inconsistencies in MRI data. Therefore, a novel hybrid framework, CE-RS-SBCIT, is introduced, integrating residual and spatial learning-based CNNs with transformer-driven modules. The proposed framework exploits local fine-grained and global contextual cues through four core innovations: (i) a smoothing and boundary-based CNN-integrated Transformer (SBCIT), (ii) tailored residual and spatial learning CNNs, (iii) a channel enhancement (CE) strategy, and (iv) a novel spatial attention mechanism. The developed SBCIT employs stem convolution and contextual interaction transformer blocks with systematic smoothing and boundary operations, enabling efficient global feature modeling. Moreover, Residual and spatial CNNs, enhanced by auxiliary transfer-learned feature maps, enrich the representation space, while the CE module amplifies discriminative channels and mitigates redundancy. Furthermore, the spatial attention mechanism selectively emphasizes subtle contrast and textural variations across tumor classes. Extensive evaluation on challenging MRI datasets from Kaggle and Figshare, encompassing glioma, meningioma, pituitary tumors, and healthy controls, demonstrates superior performance, achieving 98.30% accuracy, 98.08% sensitivity, 98.25% F1-score, and 98.43% precision.
comment: 37 Pages, 12 Figures
A Survey on Explainable Reinforcement Learning: Concepts, Algorithms, Challenges
Reinforcement Learning (RL) is a popular machine learning paradigm where intelligent agents interact with the environment to fulfill a long-term goal. Driven by the resurgence of deep learning, Deep RL (DRL) has witnessed great success over a wide spectrum of complex control tasks. Despite the encouraging results achieved, the deep neural network-based backbone is widely deemed as a black box that impedes practitioners to trust and employ trained agents in realistic scenarios where high security and reliability are essential. To alleviate this issue, a large volume of literature devoted to shedding light on the inner workings of the intelligent agents has been proposed, by constructing intrinsic interpretability or post-hoc explainability. In this survey, we provide a comprehensive review of existing works on eXplainable RL (XRL) and introduce a new taxonomy where prior works are clearly categorized into model-explaining, reward-explaining, state-explaining, and task-explaining methods. We also review and highlight RL methods that conversely leverage human knowledge to promote learning efficiency and performance of agents while this kind of method is often ignored in XRL field. Some challenges and opportunities in XRL are discussed. This survey intends to provide a high-level summarization of XRL and to motivate future research on more effective XRL solutions. Corresponding open source codes are collected and categorized at https://github.com/Plankson/awesome-explainable-reinforcement-learning.
AI Simulation by Digital Twins: Systematic Survey, Reference Framework, and Mapping to a Standardized Architecture
Insufficient data volume and quality are particularly pressing challenges in the adoption of modern subsymbolic AI. To alleviate these challenges, AI simulation uses virtual training environments in which AI agents can be safely and efficiently developed with simulated, synthetic data. Digital twins open new avenues in AI simulation, as these high-fidelity virtual replicas of physical systems are equipped with state-of-the-art simulators and the ability to further interact with the physical system for additional data collection. In this article, we report on our systematic survey of digital twin-enabled AI simulation. By analyzing 22 primary studies, we identify technological trends and derive a reference framework to situate digital twins and AI components. Based on our findings, we derive a reference framework and provide architectural guidelines by mapping it onto the ISO 23247 reference architecture for digital twins. Finally, we identify challenges and research opportunities for prospective researchers.
Towards Embodiment Scaling Laws in Robot Locomotion
Cross-embodiment generalization underpins the vision of building generalist embodied agents for any robot, yet its enabling factors remain poorly understood. We investigate embodiment scaling laws, the hypothesis that increasing the number of training embodiments improves generalization to unseen ones, using robot locomotion as a test bed. We procedurally generate ~1,000 embodiments with topological, geometric, and joint-level kinematic variations, and train policies on random subsets. We observe positive scaling trends supporting the hypothesis, and find that embodiment scaling enables substantially broader generalization than data scaling on fixed embodiments. Our best policy, trained on the full dataset, transfers zero-shot to novel embodiments in simulation and the real world, including the Unitree Go2 and H1. These results represent a step toward general embodied intelligence, with relevance to adaptive control for configurable robots, morphology co-design, and beyond.
comment: Conference on Robot Learning (CoRL), 2025. Project website: https://embodiment-scaling-laws.github.io/
Cryptography and Security 33
SCE-NTT: A Hardware Accelerator for Number Theoretic Transform Using Superconductor Electronics
This research explores the use of superconductor electronics (SCE) for accelerating fully homomorphic encryption (FHE), focusing on the Number-Theoretic Transform (NTT), a key computational bottleneck in FHE schemes. We present SCE-NTT, a dedicated hardware accelerator based on superconductive single flux quantum (SFQ) logic and memory, targeting high performance and energy efficiency beyond the limits of conventional CMOS. To address SFQ constraints such as limited dense RAM and restricted fanin/fanout, we propose a deeply pipelined NTT-128 architecture using shift register memory (SRM). Designed for N=128 32-bit coefficients, NTT-128 comprises log2(N)=7 processing elements (PEs), each featuring a butterfly unit (BU), dual coefficient memories operating in ping-pong mode via FIFO-based SRM queues, and twiddle factor buffers. The BU integrates a Shoup modular multiplier optimized for a small area, leveraging precomputed twiddle factors. A new RSFQ cell library with over 50 parameterized cells, including compound logic units, was developed for implementation. Functional and timing correctness were validated using JoSIM analog simulations and Verilog models. A multiphase clocking scheme was employed to enhance robustness and reduce path-balancing overhead, improving circuit reliability. Fabricated results show the NTT-128 unit achieves 531 million NTT/sec at 34 GHz, over 100x faster than state-of-the-art CMOS equivalents. We also project that the architecture can scale to larger sizes, such as a 2^14-point NTT in approximately 482 ns. Key-switch throughput is estimated at 1.63 million operations/sec, significantly exceeding existing hardware. These results demonstrate the strong potential of SCE-based accelerators for scalable, energy-efficient secure computation in the post-quantum era, with further gains anticipated through advances in fabrication.
comment: 13 pages, 22 figures
The WASM Cloak: Evaluating Browser Fingerprinting Defenses Under WebAssembly based Obfuscation
Browser fingerprinting defenses have historically focused on detecting JavaScript(JS)-based tracking techniques. However, the widespread adoption of WebAssembly (WASM) introduces a potential blind spot, as adversaries can convert JS to WASM's low-level binary format to obfuscate malicious logic. This paper presents the first systematic evaluation of how such WASM-based obfuscation impacts the robustness of modern fingerprinting defenses. We develop an automated pipeline that translates real-world JS fingerprinting scripts into functional WASM-obfuscated variants and test them against two classes of defenses: state-of-the-art detectors in research literature and commercial, in-browser tools. Our findings reveal a notable divergence: detectors proposed in the research literature that rely on feature-based analysis of source code show moderate vulnerability, stemming from outdated datasets or a lack of WASM compatibility. In contrast, defenses such as browser extensions and native browser features remained completely effective, as their API-level interception is agnostic to the script's underlying implementation. These results highlight a gap between academic and practical defense strategies and offer insights into strengthening detection approaches against WASM-based obfuscation, while also revealing opportunities for more evasive techniques in future attacks.
Measuring Ransomware Lateral Movement Susceptibility via Privilege-Weighted Adjacency Matrix Exponentiation
Ransomware impact hinges on how easily an intruder can move laterally and spread to the maximum number of assets. We present a graph-theoretic method to measure lateral-movement susceptibility and estimate blast radius. We build a directed multigraph where vertices represent assets and edges represent reachable services (e.g., RDP/SSH) between them. We model lateral movement as a probabilistic process using a pivot potential factor $\pi(s)$ for each service. This allows us to iteratively compute a $K$-hop compromise probability matrix that captures how compromise propagates through the network. Metrics derived from this model include: (1) Lateral-Movement Susceptibility (LMS$_K$): the average probability of a successful lateral movement between any two assets (0-1 scale); and (2) Blast-Radius Estimate (BRE$_K$): the expected percentage of assets compromised in an average attack scenario. Interactive control (SSH 22, RDP 3389) gets higher $\pi(s)$ than app-only ports (MySQL 3306, MSSQL 1433), which seldom enable pivoting without an RCE. Across anonymized enterprise snapshots, pruning high-$\pi(s)$ edges yields the largest LMS$_K$/BRE$_K$ drop, aligning with CISA guidance, MITRE ATT\&CK (TA0008: Lateral Movement), and NIST SP~800-207. The framework evaluates (micro)segmentation and helps prioritize controls that reduce lateral movement susceptibility and shrink blast radius.
comment: 14 pages
Guarding Against Malicious Biased Threats (GAMBiT) Experiments: Revealing Cognitive Bias in Human-Subjects Red-Team Cyber Range Operations
We present three large-scale human-subjects red-team cyber range datasets from the Guarding Against Malicious Biased Threats (GAMBiT) project. Across Experiments 1-3 (July 2024-March 2025), 19-20 skilled attackers per experiment conducted two 8-hour days of self-paced operations in a simulated enterprise network (SimSpace Cyber Force Platform) while we captured multi-modal data: self-reports (background, demographics, psychometrics), operational notes, terminal histories, keylogs, network packet captures (PCAP), and NIDS alerts (Suricata). Each participant began from a standardized Kali Linux VM and pursued realistic objectives (e.g., target discovery and data exfiltration) under controlled constraints. Derivative curated logs and labels are included. The combined release supports research on attacker behavior modeling, bias-aware analytics, and method benchmarking. Data are available via IEEE Dataport entries for Experiments 1-3.
Characterizing Trust Boundary Vulnerabilities in TEE Containers
Trusted Execution Environments (TEEs) have emerged as a cornerstone of confidential computing, garnering significant attention from both academia and industry. To enable the secure development, execution, and deployment, of applications on TEE platforms, TEE containers have been introduced as middleware solutions. These containers aim to shield applications from potentially malicious operating systems and orchestration interfaces while maintaining usability and reliability. In this paper, we analyze the isolation strategies employed by existing TEE containers to protect secure applications. To address the challenges in analyzing these interfaces, we designed an automated analyzer to precisely identify and evaluate their isolation boundaries. We observed that some TEE containers fail to achieve their intended goals due to critical design and implementation flaws, such as information leakage, rollback attacks, denial-of-service, and Iago attacks, which pose significant security risks. Drawing from our findings, we share key lessons to guide the development of more secure container solutions and discuss emerging trends in TEE containerization design.
PromptSleuth: Detecting Prompt Injection via Semantic Intent Invariance
Large Language Models (LLMs) are increasingly integrated into real-world applications, from virtual assistants to autonomous agents. However, their flexibility also introduces new attack vectors-particularly Prompt Injection (PI), where adversaries manipulate model behavior through crafted inputs. As attackers continuously evolve with paraphrased, obfuscated, and even multi-task injection strategies, existing benchmarks are no longer sufficient to capture the full spectrum of emerging threats. To address this gap, we construct a new benchmark that systematically extends prior efforts. Our benchmark subsumes the two widely-used existing ones while introducing new manipulation techniques and multi-task scenarios, thereby providing a more comprehensive evaluation setting. We find that existing defenses, though effective on their original benchmarks, show clear weaknesses under our benchmark, underscoring the need for more robust solutions. Our key insight is that while attack forms may vary, the adversary's intent-injecting an unauthorized task-remains invariant. Building on this observation, we propose PromptSleuth, a semantic-oriented defense framework that detects prompt injection by reasoning over task-level intent rather than surface features. Evaluated across state-of-the-art benchmarks, PromptSleuth consistently outperforms existing defense while maintaining comparable runtime and cost efficiency. These results demonstrate that intent-based semantic reasoning offers a robust, efficient, and generalizable strategy for defending LLMs against evolving prompt injection threats.
AI Agentic Vulnerability Injection And Transformation with Optimized Reasoning
The increasing complexity of software systems and the sophistication of cyber-attacks have underscored the critical need for effective automated vulnerability detection and repair systems. Traditional methods, such as static program analysis, face significant challenges related to scalability, adaptability, and high false-positive and false-negative rates. AI-driven approaches, particularly those using machine learning and deep learning models, show promise but are heavily reliant on the quality and quantity of training data. This paper introduces a novel framework designed to automatically introduce realistic, category-specific vulnerabilities into secure C/C++ codebases to generate datasets. The proposed approach coordinates multiple AI agents that simulate expert reasoning, along with function agents and traditional code analysis tools. It leverages Retrieval-Augmented Generation for contextual grounding and employs Low-Rank approximation of weights for efficient model fine-tuning. Our experimental study on 116 code samples from three different benchmarks suggests that our approach outperforms other techniques with regard to dataset accuracy, achieving between 89\% and 95\% success rates in injecting vulnerabilities at function level.
JADES: A Universal Framework for Jailbreak Assessment via Decompositional Scoring
Accurately determining whether a jailbreak attempt has succeeded is a fundamental yet unresolved challenge. Existing evaluation methods rely on misaligned proxy indicators or naive holistic judgments. They frequently misinterpret model responses, leading to inconsistent and subjective assessments that misalign with human perception. To address this gap, we introduce JADES (Jailbreak Assessment via Decompositional Scoring), a universal jailbreak evaluation framework. Its key mechanism is to automatically decompose an input harmful question into a set of weighted sub-questions, score each sub-answer, and weight-aggregate the sub-scores into a final decision. JADES also incorporates an optional fact-checking module to strengthen the detection of hallucinations in jailbreak responses. We validate JADES on JailbreakQR, a newly introduced benchmark proposed in this work, consisting of 400 pairs of jailbreak prompts and responses, each meticulously annotated by humans. In a binary setting (success/failure), JADES achieves 98.5% agreement with human evaluators, outperforming strong baselines by over 9%. Re-evaluating five popular attacks on four LLMs reveals substantial overestimation (e.g., LAA's attack success rate on GPT-3.5-Turbo drops from 93% to 69%). Our results show that JADES could deliver accurate, consistent, and interpretable evaluations, providing a reliable basis for measuring future jailbreak attacks.
comment: 17 pages, 5 figures. For the code and data supporting this work, see https://trustairlab.github.io/jades.github.io/
Multi-Agent Penetration Testing AI for the Web
AI-powered development platforms are making software creation accessible to a broader audience, but this democratization has triggered a scalability crisis in security auditing. With studies showing that up to 40% of AI-generated code contains vulnerabilities, the pace of development now vastly outstrips the capacity for thorough security assessment. We present MAPTA, a multi-agent system for autonomous web application security assessment that combines large language model orchestration with tool-grounded execution and end-to-end exploit validation. On the 104-challenge XBOW benchmark, MAPTA achieves 76.9% overall success with perfect performance on SSRF and misconfiguration vulnerabilities, 83% success on broken authorization, and strong results on injection attacks including server-side template injection (85%) and SQL injection (83%). Cross-site scripting (57%) and blind SQL injection (0%) remain challenging. Our comprehensive cost analysis across all challenges totals $21.38 with a median cost of $0.073 for successful attempts versus $0.357 for failures. Success correlates strongly with resource efficiency, enabling practical early-stopping thresholds at approximately 40 tool calls or $0.30 per challenge. MAPTA's real-world findings are impactful given both the popularity of the respective scanned GitHub repositories (8K-70K stars) and MAPTA's low average operating cost of $3.67 per open-source assessment: MAPTA discovered critical vulnerabilities including RCEs, command injections, secret exposure, and arbitrary file write vulnerabilities. Findings are responsibly disclosed, 10 findings are under CVE review.
Microarchitecture Design and Benchmarking of Custom SHA-3 Instruction for RISC-V
Integrating cryptographic accelerators into modern CPU architectures presents unique microarchitectural challenges, particularly when extending instruction sets with complex and multistage operations. Hardware-assisted cryptographic instructions, such as Intel's AES-NI and ARM's custom instructions for encryption workloads, have demonstrated substantial performance improvements. However, efficient SHA-3 acceleration remains an open problem due to its distinct permutation-based structure and memory access patterns. Existing solutions primarily rely on standalone coprocessors or software optimizations, often avoiding the complexities of direct microarchitectural integration. This study investigates the architectural challenges of embedding a SHA-3 permutation operation as a custom instruction within a general-purpose processor, focusing on pipelined simultaneous execution, storage utilization, and hardware cost. In this paper, we investigated and prototyped a SHA-3 custom instruction for the RISC-V CPU architecture. Using cycle-accurate GEM5 simulations and FPGA prototyping, our results demonstrate performance improvements of up to 8.02x for RISC-V optimized SHA-3 software workloads and up to 46.31x for Keccak-specific software workloads, with only a 15.09% increase in registers and a 11.51% increase in LUT utilization. These findings provide critical insights into the feasibility and impact of SHA-3 acceleration at the microarchitectural level, highlighting practical design considerations for future cryptographic instruction set extensions.
comment: Extended version of IEEE ISVLSI Conference Paper
CyberSleuth: Autonomous Blue-Team LLM Agent for Web Attack Forensics
Large Language Model (LLM) agents are powerful tools for automating complex tasks. In cybersecurity, researchers have primarily explored their use in red-team operations such as vulnerability discovery and penetration tests. Defensive uses for incident response and forensics have received comparatively less attention and remain at an early stage. This work presents a systematic study of LLM-agent design for the forensic investigation of realistic web application attacks. We propose CyberSleuth, an autonomous agent that processes packet-level traces and application logs to identify the targeted service, the exploited vulnerability (CVE), and attack success. We evaluate the consequences of core design decisions - spanning tool integration and agent architecture - and provide interpretable guidance for practitioners. We benchmark four agent architectures and six LLM backends on 20 incident scenarios of increasing complexity, identifying CyberSleuth as the best-performing design. In a separate set of 10 incidents from 2025, CyberSleuth correctly identifies the exact CVE in 80% of cases. At last, we conduct a human study with 22 experts, which rated the reports of CyberSleuth as complete, useful, and coherent. They also expressed a slight preference for DeepSeek R1, a good news for open source LLM. To foster progress in defensive LLM research, we release both our benchmark and the CyberSleuth platform as a foundation for fair, reproducible evaluation of forensic agents.
comment: Code: https://github.com/SmartData-Polito/LLM_Agent_Cybersecurity_Forensic
Revisiting the Privacy Risks of Split Inference: A GAN-Based Data Reconstruction Attack via Progressive Feature Optimization
The growing complexity of Deep Neural Networks (DNNs) has led to the adoption of Split Inference (SI), a collaborative paradigm that partitions computation between edge devices and the cloud to reduce latency and protect user privacy. However, recent advances in Data Reconstruction Attacks (DRAs) reveal that intermediate features exchanged in SI can be exploited to recover sensitive input data, posing significant privacy risks. Existing DRAs are typically effective only on shallow models and fail to fully leverage semantic priors, limiting their reconstruction quality and generalizability across datasets and model architectures. In this paper, we propose a novel GAN-based DRA framework with Progressive Feature Optimization (PFO), which decomposes the generator into hierarchical blocks and incrementally refines intermediate representations to enhance the semantic fidelity of reconstructed images. To stabilize the optimization and improve image realism, we introduce an L1-ball constraint during reconstruction. Extensive experiments show that our method outperforms prior attacks by a large margin, especially in high-resolution scenarios, out-of-distribution settings, and against deeper and more complex DNNs.
comment: 10 pages, 5 figures
Bitcoin as an Interplanetary Monetary Standard with Proof-of-Transit Timestamping
We explore the feasibility of deploying Bitcoin as the shared monetary standard between Earth and Mars, accounting for physical constraints of interplanetary communication. We introduce a novel primitive, Proof-of-Transit Timestamping (PoTT), to provide cryptographic, tamper-evident audit trails for Bitcoin data across high-latency, intermittently-connected links. Leveraging Delay/Disruption-Tolerant Networking (DTN) and optical low-Earth-orbit (LEO) mesh constellations, we propose an architecture for header-first replication, long-horizon Lightning channels with planetary watchtowers, and secure settlement through federated sidechains or blind-merge-mined (BMM) commit chains. We formalize PoTT, analyze its security model, and show how it measurably improves reliability and accountability without altering Bitcoin consensus or its monetary base. Near-term deployments favor strong federations for local settlement; longer-term, blind-merge-mined commit chains (if adopted) provide an alternative. The Earth L1 monetary base remains unchanged, while Mars can operate a pegged commit chain or strong federation with 1:1 pegged assets for local block production. For transparency, if both time-beacon regimes are simultaneously compromised, PoTT-M2 (and PoTT generally) reduces to administrative assertions rather than cryptographic time-anchoring.
Human-AI Collaborative Bot Detection in MMORPGs
In Massively Multiplayer Online Role-Playing Games (MMORPGs), auto-leveling bots exploit automated programs to level up characters at scale, undermining gameplay balance and fairness. Detecting such bots is challenging, not only because they mimic human behavior, but also because punitive actions require explainable justification to avoid legal and user experience issues. In this paper, we present a novel framework for detecting auto-leveling bots by leveraging contrastive representation learning and clustering techniques in a fully unsupervised manner to identify groups of characters with similar level-up patterns. To ensure reliable decisions, we incorporate a Large Language Model (LLM) as an auxiliary reviewer to validate the clustered groups, effectively mimicking a secondary human judgment. We also introduce a growth curve-based visualization to assist both the LLM and human moderators in assessing leveling behavior. This collaborative approach improves the efficiency of bot detection workflows while maintaining explainability, thereby supporting scalable and accountable bot regulation in MMORPGs.
BridgeShield: Enhancing Security for Cross-chain Bridge Applications via Heterogeneous Graph Mining
Cross-chain bridges play a vital role in enabling blockchain interoperability. However, due to the inherent design flaws and the enormous value they hold, they have become prime targets for hacker attacks. Existing detection methods show progress yet remain limited, as they mainly address single-chain behaviors and fail to capture cross-chain semantics. To address this gap, we leverage heterogeneous graph attention networks, which are well-suited for modeling multi-typed entities and relations, to capture the complex execution semantics of cross-chain behaviors. We propose BridgeShield, a detection framework that jointly models the source chain, off-chain coordination, and destination chain within a unified heterogeneous graph representation. BridgeShield incorporates intra-meta-path attention to learn fine-grained dependencies within cross-chain paths and inter-meta-path attention to highlight discriminative cross-chain patterns, thereby enabling precise identification of attack behaviors. Extensive experiments on 51 real-world cross-chain attack events demonstrate that BridgeShield achieves an average F1-score of 92.58%, representing a 24.39% improvement over state-of-the-art baselines. These results validate the effectiveness of BridgeShield as a practical solution for securing cross-chain bridges and enhancing the resilience of multi-chain ecosystems.
Enhancing Resilience for IoE: A Perspective of Networking-Level Safeguard
The Internet of Energy (IoE) integrates IoT-driven digital communication with power grids to enable efficient and sustainable energy systems. Still, its interconnectivity exposes critical infrastructure to sophisticated cyber threats, including adversarial attacks designed to bypass traditional safeguards. Unlike general IoT risks, IoE threats have heightened public safety consequences, demanding resilient solutions. From the networking-level safeguard perspective, we propose a Graph Structure Learning (GSL)-based safeguards framework that jointly optimizes graph topology and node representations to resist adversarial network model manipulation inherently. Through a conceptual overview, architectural discussion, and case study on a security dataset, we demonstrate GSL's superior robustness over representative methods, offering practitioners a viable path to secure IoE networks against evolving attacks. This work highlights the potential of GSL to enhance the resilience and reliability of future IoE networks for practitioners managing critical infrastructure. Lastly, we identify key open challenges and propose future research directions in this novel research area.
comment: To be published in IEEE Network Magazine, 2026
Evaluating Differentially Private Generation of Domain-Specific Text
Generative AI offers transformative potential for high-stakes domains such as healthcare and finance, yet privacy and regulatory barriers hinder the use of real-world data. To address this, differentially private synthetic data generation has emerged as a promising alternative. In this work, we introduce a unified benchmark to systematically evaluate the utility and fidelity of text datasets generated under formal Differential Privacy (DP) guarantees. Our benchmark addresses key challenges in domain-specific benchmarking, including choice of representative data and realistic privacy budgets, accounting for pre-training and a variety of evaluation metrics. We assess state-of-the-art privacy-preserving generation methods across five domain-specific datasets, revealing significant utility and fidelity degradation compared to real data, especially under strict privacy constraints. These findings underscore the limitations of current approaches, outline the need for advanced privacy-preserving data sharing methods and set a precedent regarding their evaluation in realistic scenarios.
Ransomware 3.0: Self-Composing and LLM-Orchestrated
Using automated reasoning, code synthesis, and contextual decision-making, we introduce a new threat that exploits large language models (LLMs) to autonomously plan, adapt, and execute the ransomware attack lifecycle. Ransomware 3.0 represents the first threat model and research prototype of LLM-orchestrated ransomware. Unlike conventional malware, the prototype only requires natural language prompts embedded in the binary; malicious code is synthesized dynamically by the LLM at runtime, yielding polymorphic variants that adapt to the execution environment. The system performs reconnaissance, payload generation, and personalized extortion, in a closed-loop attack campaign without human involvement. We evaluate this threat across personal, enterprise, and embedded environments using a phase-centric methodology that measures quantitative fidelity and qualitative coherence in each attack phase. We show that open source LLMs can generate functional ransomware components and sustain closed-loop execution across diverse environments. Finally, we present behavioral signals and multi-level telemetry of Ransomware 3.0 through a case study to motivate future development of better defenses and policy enforcements to address novel AI-enabled ransomware attacks.
Breaking Diffusion with Cache: Exploiting Approximate Caches in Diffusion Models
Diffusion models are a powerful class of generative models that produce content, such as images, from user prompts, but they are computationally intensive. To mitigate this cost, recent academic and industry work has adopted approximate caching, which reuses intermediate states from similar prompts in a cache. While efficient, this optimization introduces new security risks by breaking isolation among users. This work aims to comprehensively assess new security vulnerabilities arising from approximate caching. First, we demonstrate a remote covert channel established with the cache, where a sender injects prompts with special keywords into the cache and a receiver can recover that even after days, to exchange information. Second, we introduce a prompt stealing attack using the cache, where an attacker can recover existing cached prompts based on cache hit prompts. Finally, we introduce a poisoning attack that embeds the attacker's logos into the previously stolen prompt, to render them in future user prompts that hit the cache. These attacks are all performed remotely through the serving system, which indicates severe security vulnerabilities in approximate caching.
Federated Learning for Large Models in Medical Imaging: A Comprehensive Review
Artificial intelligence (AI) has demonstrated considerable potential in the realm of medical imaging. However, the development of high-performance AI models typically necessitates training on large-scale, centralized datasets. This approach is confronted with significant challenges due to strict patient privacy regulations and legal restrictions on data sharing and utilization. These limitations hinder the development of large-scale models in medical domains and impede continuous updates and training with new data. Federated Learning (FL), a privacy-preserving distributed training framework, offers a new solution by enabling collaborative model development across fragmented medical datasets. In this survey, we review FL's contributions at two stages of the full-stack medical analysis pipeline. First, in upstream tasks such as CT or MRI reconstruction, FL enables joint training of robust reconstruction networks on diverse, multi-institutional datasets, alleviating data scarcity while preserving confidentiality. Second, in downstream clinical tasks like tumor diagnosis and segmentation, FL supports continuous model updating by allowing local fine-tuning on new data without centralizing sensitive images. We comprehensively analyze FL implementations across the medical imaging pipeline, from physics-informed reconstruction networks to diagnostic AI systems, highlighting innovations that improve communication efficiency, align heterogeneous data, and ensure secure parameter aggregation. Meanwhile, this paper provides an outlook on future research directions, aiming to serve as a valuable reference for the field's development.
MindGuard: Tracking, Detecting, and Attributing MCP Tool Poisoning Attack via Decision Dependence Graph
The Model Context Protocol (MCP) is increasingly adopted to standardize the interaction between LLM agents and external tools. However, this trend introduces a new threat: Tool Poisoning Attacks (TPA), where tool metadata is poisoned to induce the agent to perform unauthorized operations. Existing defenses that primarily focus on behavior-level analysis are fundamentally ineffective against TPA, as poisoned tools need not be executed, leaving no behavioral trace to monitor. Thus, we propose MindGuard, a decision-level guardrail for LLM agents, providing provenance tracking of call decisions, policy-agnostic detection, and poisoning source attribution against TPA. While fully explaining LLM decision remains challenging, our empirical findings uncover a strong correlation between LLM attention mechanisms and tool invocation decisions. Therefore, we choose attention as an empirical signal for decision tracking and formalize this as the Decision Dependence Graph (DDG), which models the LLM's reasoning process as a weighted, directed graph where vertices represent logical concepts and edges quantify the attention-based dependencies. We further design robust DDG construction and graph-based anomaly analysis mechanisms that efficiently detect and attribute TPA attacks. Extensive experiments on real-world datasets demonstrate that MindGuard achieves 94\%-99\% average precision in detecting poisoned invocations, 95\%-100\% attribution accuracy, with processing times under one second and no additional token cost. Moreover, DDG can be viewed as an adaptation of the classical Program Dependence Graph (PDG), providing a solid foundation for applying traditional security policies at the decision level.
Governable AI: Provable Safety Under Extreme Threat Models
As AI rapidly advances, the security risks posed by AI are becoming increasingly severe, especially in critical scenarios, including those posing existential risks. If AI becomes uncontrollable, manipulated, or actively evades safety mechanisms, it could trigger systemic disasters. Existing AI safety approaches-such as model enhancement, value alignment, and human intervention-suffer from fundamental, in-principle limitations when facing AI with extreme motivations and unlimited intelligence, and cannot guarantee security. To address this challenge, we propose a Governable AI (GAI) framework that shifts from traditional internal constraints to externally enforced structural compliance based on cryptographic mechanisms that are computationally infeasible to break, even for future AI, under the defined threat model and well-established cryptographic assumptions.The GAI framework is composed of a simple yet reliable, fully deterministic, powerful, flexible, and general-purpose rule enforcement module (REM); governance rules; and a governable secure super-platform (GSSP) that offers end-to-end protection against compromise or subversion by AI. The decoupling of the governance rules and the technical platform further enables a feasible and generalizable technical pathway for the safety governance of AI. REM enforces the bottom line defined by governance rules, while GSSP ensures non-bypassability, tamper-resistance, and unforgeability to eliminate all identified attack vectors. This paper also presents a rigorous formal proof of the security properties of this mechanism and demonstrates its effectiveness through a prototype implementation evaluated in representative high-stakes scenarios.
Safe-Control: A Safety Patch for Mitigating Unsafe Content in Text-to-Image Generation Models
Despite the advancements in Text-to-Image (T2I) generation models, their potential for misuse or even abuse raises serious safety concerns. Model developers have made tremendous efforts to introduce safety mechanisms that can address these concerns in T2I models. However, the existing safety mechanisms, whether external or internal, either remain susceptible to evasion under distribution shifts or require extensive model-specific adjustments. To address these limitations, we introduce Safe-Control, an innovative plug-and-play safety patch designed to mitigate unsafe content generation in T2I models. Using data-driven strategies and safety-aware conditions, Safe-Control injects safety control signals into the locked T2I model, acting as an update in a patch-like manner. Model developers can also construct various safety patches to meet the evolving safety requirements, which can be flexibly merged into a single, unified patch. Its plug-and-play design further ensures adaptability, making it compatible with other T2I models of similar denoising architecture. We conduct extensive evaluations on six diverse and public T2I models. Empirical results highlight that Safe-Control is effective in reducing unsafe content generation across six diverse T2I models with similar generative architectures, yet it successfully maintains the quality and text alignment of benign images. Compared to seven state-of-the-art safety mechanisms, including both external and internal defenses, Safe-Control significantly outperforms all baselines in reducing unsafe content generation. For example, it reduces the probability of unsafe content generation to 7%, compared to approximately 20% for most baseline methods, under both unsafe prompts and the latest adversarial attacks.
Improving Google A2A Protocol: Protecting Sensitive Data and Mitigating Unintended Harms in Multi-Agent Systems
Googles A2A protocol provides a secure communication framework for AI agents but demonstrates critical limitations when handling highly sensitive information such as payment credentials and identity documents. These gaps increase the risk of unintended harms, including unauthorized disclosure, privilege escalation, and misuse of private data in generative multi-agent environments. In this paper, we identify key weaknesses of A2A: insufficient token lifetime control, lack of strong customer authentication, overbroad access scopes, and missing consent flows. We propose protocol-level enhancements grounded in a structured threat model for semi-trusted multi-agent systems. Our refinements introduce explicit consent orchestration, ephemeral scoped tokens, and direct user-to-service data channels to minimize exposure across time, context, and topology. Empirical evaluation using adversarial prompt injection tests shows that the enhanced protocol substantially reduces sensitive data leakage while maintaining low communication latency. Comparative analysis highlights the advantages of our approach over both the original A2A specification and related academic proposals. These contributions establish a practical path for evolving A2A into a privacy-preserving framework that mitigates unintended harms in multi-agent generative AI systems.
DPImageBench: A Unified Benchmark for Differentially Private Image Synthesis CCS 2025
Differentially private (DP) image synthesis aims to generate artificial images that retain the properties of sensitive images while protecting the privacy of individual images within the dataset. Despite recent advancements, we find that inconsistent--and sometimes flawed--evaluation protocols have been applied across studies. This not only impedes the understanding of current methods but also hinders future advancements. To address the issue, this paper introduces DPImageBench for DP image synthesis, with thoughtful design across several dimensions: (1) Methods. We study eleven prominent methods and systematically characterize each based on model architecture, pretraining strategy, and privacy mechanism. (2) Evaluation. We include nine datasets and seven fidelity and utility metrics to thoroughly assess them. Notably, we find that a common practice of selecting downstream classifiers based on the highest accuracy on the sensitive test set not only violates DP but also overestimates the utility scores. DPImageBench corrects for these mistakes. (3) Platform. Despite the methods and evaluation protocols, DPImageBench provides a standardized interface that accommodates current and future implementations within a unified framework. With DPImageBench, we have several noteworthy findings. For example, contrary to the common wisdom that pretraining on public image datasets is usually beneficial, we find that the distributional similarity between pretraining and sensitive images significantly impacts the performance of the synthetic images and does not always yield improvements. In addition, adding noise to low-dimensional features, such as the high-level characteristics of sensitive images, is less affected by the privacy budget compared to adding noise to high-dimensional features, like weight gradients. The former methods perform better than the latter under a low privacy budget.
comment: Accepted at CCS 2025; The first two authors contributed equally; code available at https://github.com/2019ChenGong/DPImageBench
Federated Distributed Key Generation
Distributed Key Generation (DKG) is vital to threshold-based cryptographic protocols such as threshold signatures, secure multiparty computation, and i-voting. Yet, standard $(n,t)$-DKG requires a known set of $n$ participants and a fixed threshold $t$, making it impractical for public or decentralized settings where membership and availability can change. We introduce Federated Distributed Key Generation (FDKG), which relaxes these constraints by allowing each participant to select its own guardian set, with a local threshold to reconstruct that participant's partial key. FDKG generalizes DKG and draws inspiration from Federated Byzantine Agreement, enabling dynamic trust delegation with minimal message complexity (two rounds). The protocol's liveness can tolerate adversary that controls up to $k - t + 1$ nodes in every guardian set. The paper presents a detailed protocol, a formal description of liveness, privacy, and integrity properties, and a simulation-based evaluation showcasing the efficacy of FDKG in mitigating node unreliability. In a setting of 100 parties, a 50% participation rate, 80% retention, and 40 guardians, the distribution phase incurred a total message size of 332.7 kB ($O(n\,k)$), and reconstruction phase 416.56 kB ($O(n\,k)$. Groth16 client-side proving took about 5 s in the distribution phase and ranged from 0.619 s up to 29.619 s in the reconstruction phase. Our work advances distributed cryptography by enabling flexible trust models for dynamic networks, with applications ranging from ad-hoc collaboration to blockchain governance.
ADAGE: Active Defenses Against GNN Extraction
Graph Neural Networks (GNNs) achieve high performance in various real-world applications, such as drug discovery, traffic states prediction, and recommendation systems. The fact that building powerful GNNs requires a large amount of training data, powerful computing resources, and human expertise turns the models into lucrative targets for model stealing attacks. Prior work has revealed that the threat vector of stealing attacks against GNNs is large and diverse, as an attacker can leverage various heterogeneous signals ranging from node labels to high-dimensional node embeddings to create a local copy of the target GNN at a fraction of the original training costs. This diversity in the threat vector renders the design of effective and general defenses challenging and existing defenses usually focus on one particular stealing setup. Additionally, they solely provide means to identify stolen model copies rather than preventing the attack. To close this gap, we propose the first and general Active Defense Against GNN Extraction (ADAGE). ADAGE builds on the observation that stealing a model's full functionality requires highly diverse queries to leak its behavior across the input space. Our defense monitors this query diversity and progressively perturbs outputs as the accumulated leakage grows. In contrast to prior work, ADAGE can prevent stealing across all common attack setups. Our extensive experimental evaluation using six benchmark datasets, four GNN models, and three types of adaptive attackers shows that ADAGE penalizes attackers to the degree of rendering stealing impossible, whilst preserving predictive performance on downstream tasks. ADAGE, thereby, contributes towards securely sharing valuable GNNs in the future.
comment: Not all authors have given their explicit consent
Maven-Hijack: Software Supply Chain Attack Exploiting Packaging Order
Java projects frequently rely on package managers such as Maven to manage complex webs of external dependencies. While these tools streamline development, they also introduce subtle risks to the software supply chain. In this paper, we present Maven-Hijack, a novel attack that exploits the order in which Maven packages dependencies and the way the Java Virtual Machine resolves classes at runtime. By injecting a malicious class with the same fully qualified name as a legitimate one into a dependency that is packaged earlier, an attacker can silently override core application behavior without modifying the main codebase or library names. We demonstrate the real-world feasibility of this attack by compromising the Corona-Warn-App, a widely used open-source COVID-19 contact tracing system, and gaining control over its database connection logic. We evaluate three mitigation strategies, such as sealed JARs, Java Modules, and the Maven Enforcer plugin. Our results show that, while Java Modules offer strong protection, the Maven Enforcer plugin with duplicate class detection provides the most practical and effective defense for current Java projects. These findings highlight the urgent need for improved safeguards in Java's build and dependency management processes to prevent stealthy supply chain attacks.
comment: 8 pages, added a mitigation chapter
Privacy-Aware Detection of Fake Identity Documents: Methodology, Benchmark, and Improved Algorithms (FakeIDet2)
Remote user verification in Internet-based applications is becoming increasingly important nowadays. A popular scenario for it consists of submitting a picture of the user's Identity Document (ID) to a service platform, authenticating its veracity, and then granting access to the requested digital service. An ID is well-suited to verify the identity of an individual, since it is government issued, unique, and nontransferable. However, with recent advances in Artificial Intelligence (AI), attackers can surpass security measures in IDs and create very realistic physical and synthetic fake IDs. Researchers are now trying to develop methods to detect an ever-growing number of these AI-based fakes that are almost indistinguishable from authentic (bona fide) IDs. In this counterattack effort, researchers are faced with an important challenge: the difficulty in using real data to train fake ID detectors. This real data scarcity for research and development is originated by the sensitive nature of these documents, which are usually kept private by the ID owners (the users) and the ID Holders (e.g., government, police, bank, etc.). The main contributions of our study are: 1) We propose and discuss a patch-based methodology to preserve privacy in fake ID detection research. 2) We provide a new public database, FakeIDet2-db, comprising over 900K real/fake ID patches extracted from 2,000 ID images, acquired using different smartphone sensors, illumination and height conditions, etc. In addition, three physical attacks are considered: print, screen, and composite. 3) We present a new privacy-aware fake ID detection method, FakeIDet2. 4) We release a standard reproducible benchmark that considers physical and synthetic attacks from popular databases in the literature.
Probabilistic Modeling of Jailbreak on Multimodal LLMs: From Quantification to Application
Recently, Multimodal Large Language Models (MLLMs) have demonstrated their superior ability in understanding multimodal content. However, they remain vulnerable to jailbreak attacks, which exploit weaknesses in their safety alignment to generate harmful responses. Previous studies categorize jailbreaks as successful or failed based on whether responses contain malicious content. However, given the stochastic nature of MLLM responses, this binary classification of an input's ability to jailbreak MLLMs is inappropriate. Derived from this viewpoint, we introduce jailbreak probability to quantify the jailbreak potential of an input, which represents the likelihood that MLLMs generated a malicious response when prompted with this input. We approximate this probability through multiple queries to MLLMs. After modeling the relationship between input hidden states and their corresponding jailbreak probability using Jailbreak Probability Prediction Network (JPPN), we use continuous jailbreak probability for optimization. Specifically, we propose Jailbreak-Probability-based Attack (JPA) that optimizes adversarial perturbations on input image to maximize jailbreak probability, and further enhance it as Multimodal JPA (MJPA) by including monotonic text rephrasing. To counteract attacks, we also propose Jailbreak-Probability-based Finetuning (JPF), which minimizes jailbreak probability through MLLM parameter updates. Extensive experiments show that (1) (M)JPA yields significant improvements when attacking a wide range of models under both white and black box settings. (2) JPF vastly reduces jailbreaks by at most over 60\%. Both of the above results demonstrate the significance of introducing jailbreak probability to make nuanced distinctions among input jailbreak abilities.
Formal Verification of Physical Layer Security Protocols for Next-Generation Communication Networks (extended version)
Formal verification is crucial for ensuring the robustness of security protocols against adversarial attacks. The Needham-Schroeder protocol, a foundational authentication mechanism, has been extensively studied, including its integration with Physical Layer Security (PLS) techniques such as watermarking and jamming. Recent research has used ProVerif to verify these mechanisms in terms of secrecy. However, the ProVerif-based approach limits the ability to improve understanding of security beyond verification results. To overcome these limitations, we re-model the same protocol using an Isabelle formalism that generates sound animation, enabling interactive and automated formal verification of security protocols. Our modelling and verification framework is generic and highly configurable, supporting both cryptography and PLS. For the same protocol, we have conducted a comprehensive analysis (secrecy and authenticity in four different eavesdropper locations under both passive and active attacks) using our new web interface. Our findings not only successfully reproduce and reinforce previous results on secrecy but also reveal an uncommon but expected outcome: authenticity is preserved across all examined scenarios, even in cases where secrecy is compromised. We have proposed a PLS-based Diffie-Hellman protocol that integrates watermarking and jamming, and our analysis shows that it is secure for deriving a session key with required authentication. These highlight the advantages of our novel approach, demonstrating its robustness in formally verifying security properties beyond conventional methods.
comment: Extended version (with appendices) of the camera-ready for ICFEM2025; 24 pages, 3 tables, and 6 figures
When Memory Becomes a Vulnerability: Towards Multi-turn Jailbreak Attacks against Text-to-Image Generation Systems
Modern text-to-image (T2I) generation systems (e.g., DALL$\cdot$E 3) exploit the memory mechanism, which captures key information in multi-turn interactions for faithful generation. Despite its practicality, the security analyses of this mechanism have fallen far behind. In this paper, we reveal that it can exacerbate the risk of jailbreak attacks. Previous attacks fuse the unsafe target prompt into one ultimate adversarial prompt, which can be easily detected or lead to the generation of non-unsafe images due to under- or over-detoxification. In contrast, we propose embedding the malice at the inception of the chat session in memory, addressing the above limitations. Specifically, we propose Inception, the first multi-turn jailbreak attack against real-world text-to-image generation systems that explicitly exploits their memory mechanisms. Inception is composed of two key modules: segmentation and recursion. We introduce Segmentation, a semantic-preserving method that generates multi-round prompts. By leveraging NLP analysis techniques, we design policies to decompose a prompt, together with its malicious intent, according to sentence structure, thereby evading safety filters. Recursion further addresses the challenge posed by unsafe sub-prompts that cannot be separated through simple segmentation. It firstly expands the sub-prompt, then invokes segmentation recursively. To facilitate multi-turn adversarial prompts crafting, we build VisionFlow, an emulation T2I system that integrates two-stage safety filters and industrial-grade memory mechanisms. The experiment results show that Inception successfully allures unsafe image generation, surpassing the SOTA by a 20.0\% margin in attack success rate. We also conduct experiments on the real-world commercial T2I generation platforms, further validating the threats of Inception in practice.
comment: This work proposes a multi-turn jailbreak attack against real-world chat-based T2I generation systems that intergrate memory mechanism. It also constructed a simulation system, with considering three industrial-grade memory mechanisms, 7 kinds of safety filters (both input and output)
A Time Series Analysis of Malware Uploads to Programming Language Ecosystems
Software ecosystems built around programming languages have greatly facilitated software development. At the same time, their security has increasingly been acknowledged as a problem. To this end, the paper examines the previously overlooked longitudinal aspects of software ecosystem security, focusing on malware uploaded to six popular programming language ecosystems. The dataset examined is based on the new Open Source Vulnerabilities (OSV) database. According to the results, records about detected malware uploads in the database have recently surpassed those addressing vulnerabilities in packages distributed in the ecosystems. In the early 2025 even up to 80% of all entries in the OSV have been about malware. Regarding time series analysis of malware frequencies and their shares to all database entries, good predictions are available already by relatively simple autoregressive models using the numbers of ecosystems, security advisories, and media and other articles as predictors. With these results and the accompanying discussion, the paper improves and advances the understanding of the thus far overlooked longitudinal aspects of ecosystems and malware.
comment: Proceedings of the 20th International Conference on Availability, Reliability and Security (ARES 2025), Ghent, Springer, pp. 269-285. Please note that this version diverges from the publisher's definite version. A new version will be uploaded once the publisher's embargo period is over
Artificial Intelligence 20
Deep Active Learning for Lung Disease Severity Classification from Chest X-rays: Learning with Less Data in the Presence of Class Imbalance
To reduce the amount of required labeled data for lung disease severity classification from chest X-rays (CXRs) under class imbalance, this study applied deep active learning with a Bayesian Neural Network (BNN) approximation and weighted loss function. This retrospective study collected 2,319 CXRs from 963 patients (mean age, 59.2 $\pm$ 16.6 years; 481 female) at Emory Healthcare affiliated hospitals between January and November 2020. All patients had clinically confirmed COVID-19. Each CXR was independently labeled by 3 to 6 board-certified radiologists as normal, moderate, or severe. A deep neural network with Monte Carlo Dropout was trained using active learning to classify disease severity. Various acquisition functions were used to iteratively select the most informative samples from an unlabeled pool. Performance was evaluated using accuracy, area under the receiver operating characteristic curve (AU ROC), and area under the precision-recall curve (AU PRC). Training time and acquisition time were recorded. Statistical analysis included descriptive metrics and performance comparisons across acquisition strategies. Entropy Sampling achieved 93.7% accuracy (AU ROC, 0.91) in binary classification (normal vs. diseased) using 15.4% of the training data. In the multi-class setting, Mean STD sampling achieved 70.3% accuracy (AU ROC, 0.86) using 23.1% of the labeled data. These methods outperformed more complex and computationally expensive acquisition functions and significantly reduced labeling needs. Deep active learning with BNN approximation and weighted loss effectively reduces labeled data requirements while addressing class imbalance, maintaining or exceeding diagnostic performance.
Breaking the Cold-Start Barrier: Reinforcement Learning with Double and Dueling DQNs
Recommender systems struggle to provide accurate suggestions to new users with limited interaction history, a challenge known as the cold-user problem. This paper proposes a reinforcement learning approach using Double and Dueling Deep Q-Networks (DQN) to dynamically learn user preferences from sparse feedback, enhancing recommendation accuracy without relying on sensitive demographic data. By integrating these advanced DQN variants with a matrix factorization model, we achieve superior performance on a large e-commerce dataset compared to traditional methods like popularity-based and active learning strategies. Experimental results show that our method, particularly Dueling DQN, reduces Root Mean Square Error (RMSE) for cold users, offering an effective solution for privacy-constrained environments.
Reinforcement Learning for Optimizing Large Qubit Array based Quantum Sensor Circuits
As the number of qubits in a sensor increases, the complexity of designing and controlling the quantum circuits grows exponentially. Manually optimizing these circuits becomes infeasible. Optimizing entanglement distribution in large-scale quantum circuits is critical for enhancing the sensitivity and efficiency of quantum sensors [5], [6]. This paper presents an engineering integration of reinforcement learning with tensor-network-based simulation (MPS) for scalable circuit optimization for optimizing quantum sensor circuits with up to 60 qubits. To enable efficient simulation and scalability, we adopt tensor network methods, specifically the Matrix Product State (MPS) representation, instead of traditional state vector or density matrix approaches. Our reinforcement learning agent learns to restructure circuits to maximize Quantum Fisher Information (QFI) and entanglement entropy while reducing gate counts and circuit depth. Experimental results show consistent improvements, with QFI values approaching 1, entanglement entropy in the 0.8-1.0 range, and up to 90% reduction in depth and gate count. These results highlight the potential of combining quantum machine learning and tensor networks to optimize complex quantum circuits under realistic constraints.
comment: 10 pages, 13 figures, 2 tables
Quantum Machine Learning for Optimizing Entanglement Distribution in Quantum Sensor Circuits
In the rapidly evolving field of quantum computing, optimizing quantum circuits for specific tasks is crucial for enhancing performance and efficiency. More recently, quantum sensing has become a distinct and rapidly growing branch of research within the area of quantum science and technology. The field is expected to provide new opportunities, especially regarding high sensitivity and precision. Entanglement is one of the key factors in achieving high sensitivity and measurement precision [3]. This paper presents a novel approach utilizing quantum machine learning techniques to optimize entanglement distribution in quantum sensor circuits. By leveraging reinforcement learning within a quantum environment, we aim to optimize the entanglement layout to maximize Quantum Fisher Information (QFI) and entanglement entropy, which are key indicators of a quantum system's sensitivity and coherence, while minimizing circuit depth and gate counts. Our implementation, based on Qiskit, integrates noise models and error mitigation strategies to simulate realistic quantum environments. The results demonstrate significant improvements in circuit performance and sensitivity, highlighting the potential of machine learning in quantum circuit optimization by measuring high QFI and entropy in the range of 0.84-1.0 with depth and gate count reduction by 20-86%.
comment: 11 pages, 13 figures, 4 tables
A Mixture of Experts Gating Network for Enhanced Surrogate Modeling in External Aerodynamics
The computational cost associated with high-fidelity CFD simulations remains a significant bottleneck in the automotive design and optimization cycle. While ML-based surrogate models have emerged as a promising alternative to accelerate aerodynamic predictions, the field is characterized by a diverse and rapidly evolving landscape of specialized neural network architectures, with no single model demonstrating universal superiority. This paper introduces a novel meta-learning framework that leverages this architectural diversity as a strength. We propose a Mixture of Experts (MoE) model that employs a dedicated gating network to dynamically and optimally combine the predictions from three heterogeneous, state-of-the-art surrogate models: DoMINO, a decomposable multi-scale neural operator; X-MeshGraphNet, a scalable multi-scale graph neural network; and FigConvNet, a factorized implicit global convolution network. The gating network learns a spatially-variant weighting strategy, assigning credibility to each expert based on its localized performance in predicting surface pressure and wall shear stress fields. To prevent model collapse and encourage balanced expert contributions, we integrate an entropy regularization term into the training loss function. The entire system is trained and validated on the DrivAerML dataset, a large-scale, public benchmark of high-fidelity CFD simulations for automotive aerodynamics. Quantitative results demonstrate that the MoE model achieves a significant reduction in L-2 prediction error, outperforming not only the ensemble average but also the most accurate individual expert model across all evaluated physical quantities. This work establishes the MoE framework as a powerful and effective strategy for creating more robust and accurate composite surrogate models by synergistically combining the complementary strengths of specialized architectures.
Zero-Shot KWS for Children's Speech using Layer-Wise Features from SSL Models
Numerous methods have been proposed to enhance Keyword Spotting (KWS) in adult speech, but children's speech presents unique challenges for KWS systems due to its distinct acoustic and linguistic characteristics. This paper introduces a zero-shot KWS approach that leverages state-of-the-art self-supervised learning (SSL) models, including Wav2Vec2, HuBERT and Data2Vec. Features are extracted layer-wise from these SSL models and used to train a Kaldi-based DNN KWS system. The WSJCAM0 adult speech dataset was used for training, while the PFSTAR children's speech dataset was used for testing, demonstrating the zero-shot capability of our method. Our approach achieved state-of-the-art results across all keyword sets for children's speech. Notably, the Wav2Vec2 model, particularly layer 22, performed the best, delivering an ATWV score of 0.691, a MTWV score of 0.7003 and probability of false alarm and probability of miss of 0.0164 and 0.0547 respectively, for a set of 30 keywords. Furthermore, age-specific performance evaluation confirmed the system's effectiveness across different age groups of children. To assess the system's robustness against noise, additional experiments were conducted using the best-performing layer of the best-performing Wav2Vec2 model. The results demonstrated a significant improvement over traditional MFCC-based baseline, emphasizing the potential of SSL embeddings even in noisy conditions. To further generalize the KWS framework, the experiments were repeated for an additional CMU dataset. Overall the results highlight the significant contribution of SSL features in enhancing Zero-Shot KWS performance for children's speech, effectively addressing the challenges associated with the distinct characteristics of child speakers.
comment: Accepted
HCQA: Hybrid Classical-Quantum Agent for Generating Optimal Quantum Sensor Circuits
This study proposes an HCQA for designing optimal Quantum Sensor Circuits (QSCs) to address complex quantum physics problems. The HCQA integrates computational intelligence techniques by leveraging a Deep Q-Network (DQN) for learning and policy optimization, enhanced by a quantum-based action selection mechanism based on the Q-values. A quantum circuit encodes the agent current state using Ry gates, and then creates a superposition of possible actions. Measurement of the circuit results in probabilistic action outcomes, allowing the agent to generate optimal QSCs by selecting sequences of gates that maximize the Quantum Fisher Information (QFI) while minimizing the number of gates. This computational intelligence-driven HCQA enables the automated generation of entangled quantum states, specifically the squeezed states, with high QFI sensitivity for quantum state estimation and control. Evaluation of the HCQA on a QSC that consists of two qubits and a sequence of Rx, Ry, and S gates demonstrates its efficiency in generating optimal QSCs with a QFI of 1. This work highlights the synergy between AI-driven learning and quantum computation, illustrating how intelligent agents can autonomously discover optimal quantum circuit designs for enhanced sensing and estimation tasks.
comment: 9 pages, 9 figures
Full-Frequency Temporal Patching and Structured Masking for Enhanced Audio Classification
Transformers and State-Space Models (SSMs) have advanced audio classification by modeling spectrograms as sequences of patches. However, existing models such as the Audio Spectrogram Transformer (AST) and Audio Mamba (AuM) adopt square patching from computer vision, which disrupts continuous frequency patterns and produces an excessive number of patches, slowing training, and increasing computation. We propose Full-Frequency Temporal Patching (FFTP), a patching strategy that better matches the time-frequency asymmetry of spectrograms by spanning full frequency bands with localized temporal context, preserving harmonic structure, and significantly reducing patch count and computation. We also introduce SpecMask, a patch-aligned spectrogram augmentation that combines full-frequency and localized time-frequency masks under a fixed masking budget, enhancing temporal robustness while preserving spectral continuity. When applied on both AST and AuM, our patching method with SpecMask improves mAP by up to +6.76 on AudioSet-18k and accuracy by up to +8.46 on SpeechCommandsV2, while reducing computation by up to 83.26%, demonstrating both performance and efficiency gains.
Addressing accuracy and hallucination of LLMs in Alzheimer's disease research through knowledge graphs
In the past two years, large language model (LLM)-based chatbots, such as ChatGPT, have revolutionized various domains by enabling diverse task completion and question-answering capabilities. However, their application in scientific research remains constrained by challenges such as hallucinations, limited domain-specific knowledge, and lack of explainability or traceability for the response. Graph-based Retrieval-Augmented Generation (GraphRAG) has emerged as a promising approach to improving chatbot reliability by integrating domain-specific contextual information before response generation, addressing some limitations of standard LLMs. Despite its potential, there are only limited studies that evaluate GraphRAG on specific domains that require intensive knowledge, like Alzheimer's disease or other biomedical domains. In this paper, we assess the quality and traceability of two popular GraphRAG systems. We compile a database of 50 papers and 70 expert questions related to Alzheimer's disease, construct a GraphRAG knowledge base, and employ GPT-4o as the LLM for answering queries. We then compare the quality of responses generated by GraphRAG with those from a standard GPT-4o model. Additionally, we discuss and evaluate the traceability of several Retrieval-Augmented Generation (RAG) and GraphRAG systems. Finally, we provide an easy-to-use interface with a pre-built Alzheimer's disease database for researchers to test the performance of both standard RAG and GraphRAG.
Decoding Memories: An Efficient Pipeline for Self-Consistency Hallucination Detection
Large language models (LLMs) have demonstrated impressive performance in both research and real-world applications, but they still struggle with hallucination. Existing hallucination detection methods often perform poorly on sentence-level generation or rely heavily on domain-specific knowledge. While self-consistency approaches help address these limitations, they incur high computational costs due to repeated generation. In this paper, we conduct the first study on identifying redundancy in self-consistency methods, manifested as shared prefix tokens across generations, and observe that non-exact-answer tokens contribute minimally to the semantic content. Based on these insights, we propose a novel Decoding Memory Pipeline (DMP) that accelerates generation through selective inference and annealed decoding. Being orthogonal to the model, dataset, decoding strategy, and self-consistency baseline, our DMP consistently improves the efficiency of multi-response generation and holds promise for extension to alignment and reasoning tasks. Extensive experiments show that our method achieves up to a 3x speedup without sacrificing AUROC performance.
comment: 14 pages, under review
Can Layer-wise SSL Features Improve Zero-Shot ASR Performance for Children's Speech?
Automatic Speech Recognition (ASR) systems often struggle to accurately process children's speech due to its distinct and highly variable acoustic and linguistic characteristics. While recent advancements in self-supervised learning (SSL) models have greatly enhanced the transcription of adult speech, accurately transcribing children's speech remains a significant challenge. This study investigates the effectiveness of layer-wise features extracted from state-of-the-art SSL pre-trained models - specifically, Wav2Vec2, HuBERT, Data2Vec, and WavLM in improving the performance of ASR for children's speech in zero-shot scenarios. A detailed analysis of features extracted from these models was conducted, integrating them into a simplified DNN-based ASR system using the Kaldi toolkit. The analysis identified the most effective layers for enhancing ASR performance on children's speech in a zero-shot scenario, where WSJCAM0 adult speech was used for training and PFSTAR children speech for testing. Experimental results indicated that Layer 22 of the Wav2Vec2 model achieved the lowest Word Error Rate (WER) of 5.15%, representing a 51.64% relative improvement over the direct zero-shot decoding using Wav2Vec2 (WER of 10.65%). Additionally, age group-wise analysis demonstrated consistent performance improvements with increasing age, along with significant gains observed even in younger age groups using the SSL features. Further experiments on the CMU Kids dataset confirmed similar trends, highlighting the generalizability of the proposed approach.
comment: Accepted
Generalizable Object Re-Identification via Visual In-Context Prompting ICCV 2025
Current object re-identification (ReID) methods train domain-specific models (e.g., for persons or vehicles), which lack generalization and demand costly labeled data for new categories. While self-supervised learning reduces annotation needs by learning instance-wise invariance, it struggles to capture \textit{identity-sensitive} features critical for ReID. This paper proposes Visual In-Context Prompting~(VICP), a novel framework where models trained on seen categories can directly generalize to unseen novel categories using only \textit{in-context examples} as prompts, without requiring parameter adaptation. VICP synergizes LLMs and vision foundation models~(VFM): LLMs infer semantic identity rules from few-shot positive/negative pairs through task-specific prompting, which then guides a VFM (\eg, DINO) to extract ID-discriminative features via \textit{dynamic visual prompts}. By aligning LLM-derived semantic concepts with the VFM's pre-trained prior, VICP enables generalization to novel categories, eliminating the need for dataset-specific retraining. To support evaluation, we introduce ShopID10K, a dataset of 10K object instances from e-commerce platforms, featuring multi-view images and cross-domain testing. Experiments on ShopID10K and diverse ReID benchmarks demonstrate that VICP outperforms baselines by a clear margin on unseen categories. Code is available at https://github.com/Hzzone/VICP.
comment: ICCV 2025
Enhancing Robustness of Autoregressive Language Models against Orthographic Attacks via Pixel-based Approach
Autoregressive language models are vulnerable to orthographic attacks, where input text is perturbed with characters from multilingual alphabets, leading to substantial performance degradation. This vulnerability primarily stems from the out-of-vocabulary issue inherent in subword tokenizers and their embeddings. To address this limitation, we propose a pixel-based generative language model that replaces the text-based embeddings with pixel-based representations by rendering words as individual images. This design provides stronger robustness to noisy inputs, while an extension of compatibility to multilingual text across diverse writing systems. We evaluate the proposed method on the multilingual LAMBADA dataset, WMT24 dataset and the SST-2 benchmark, demonstrating both its resilience to orthographic noise and its effectiveness in multilingual settings.
Fuzzy, Symbolic, and Contextual: Enhancing LLM Instruction via Cognitive Scaffolding
We study how architectural inductive biases influence the cognitive behavior of large language models (LLMs) in instructional dialogue. We introduce a symbolic scaffolding mechanism paired with a short-term memory schema designed to promote adaptive, structured reasoning in Socratic tutoring. Using controlled ablation across five system variants, we evaluate model outputs via expert-designed rubrics covering scaffolding, responsiveness, symbolic reasoning, and conversational memory. We present preliminary results using an LLM-based evaluation framework aligned to a cognitively grounded rubric. This enables scalable, systematic comparisons across architectural variants in early-stage experimentation. The preliminary results show that our full system consistently outperforms baseline variants. Analysis reveals that removing memory or symbolic structure degrades key cognitive behaviors, including abstraction, adaptive probing, and conceptual continuity. These findings support a processing-level account in which architectural scaffolds can reliably shape emergent instructional strategies in LLMs.
Improving Aviation Safety Analysis: Automated HFACS Classification Using Reinforcement Learning with Group Relative Policy Optimization
Analyzing the human factors behind aviation accidents is crucial for preventing future incidents, yet traditional methods using the Human Factors Analysis and Classification System (HFACS) are limited by scalability and consistency. To address this, we introduce an automated HFACS classification framework for aviation safety analysis that utilizes Reinforcement Learning with Group Relative Policy Optimization (GRPO) to fine-tune a Llama-3.1 8B language model. Our approach incorporates a multi-component reward system tailored for aviation safety analysis and integrates synthetic data generation to overcome class imbalance in accident datasets. The resulting GRPO-optimized model achieved noticeable performance gains, including a 350% increase in exact match accuracy (from 0.0400 to 0.1800) and an improved partial match accuracy of 0.8800. Significantly, our specialized model outperforms state-of-the-art LLMs (Large Language Models), including GPT-5-mini and Gemini-2.5-fiash, on key metrics. This research also proposes exact match accuracy in multi-label HFACS classification problem as a new benchmarking methodology to evaluate the advanced reasoning capabilities of language models. Ultimately, our work validates that smaller, domain-optimized models can provide a computationally efficient and better solution for critical safety analysis. This approach makes powerful, low-latency deployment on resource-constrained edge devices feasible.
LLM Test Generation via Iterative Hybrid Program Analysis ICSE 2026
Automating unit test generation remains a significant challenge, particularly for complex methods in real-world projects. While Large Language Models (LLMs) have made strides in code generation, they struggle to achieve high branch coverage due to their limited ability to reason about intricate control flow structures. To address this limitation, we introduce Panta, a technique that emulates the iterative process human developers follow when analyzing code and constructing test cases. Panta integrates static control flow analysis and dynamic code coverage analysis to systematically guide LLMs in identifying uncovered execution paths and generating better test cases. By incorporating an iterative feedback-driven mechanism, our technique continuously refines test generation based on static and dynamic path coverage insights, ensuring more comprehensive and effective testing. Our empirical evaluation, conducted on classes with high cyclomatic complexity from open-source projects, demonstrates that Panta achieves 26% higher line coverage and 23% higher branch coverage compared to the state-of-the-art.
comment: This paper has been accepted for publication in 2026 IEEE/ACM 48th International Conference on Software Engineering (ICSE 2026)
FrontierMath: A Benchmark for Evaluating Advanced Mathematical Reasoning in AI
We introduce FrontierMath, a benchmark of hundreds of original, exceptionally challenging mathematics problems crafted and vetted by expert mathematicians. The questions cover most major branches of modern mathematics -- from computationally intensive problems in number theory and real analysis to abstract questions in algebraic geometry and category theory. Solving a typical problem requires multiple hours of effort from a researcher in the relevant branch of mathematics, and for the upper end questions, multiple days. FrontierMath uses new, unpublished problems and automated verification to reliably evaluate models while minimizing risk of data contamination. Current state-of-the-art AI models solve under 2% of problems, revealing a vast gap between AI capabilities and the prowess of the mathematical community. As AI systems advance toward expert-level mathematical abilities, FrontierMath offers a rigorous testbed that quantifies their progress.
PlantVillageVQA: A Visual Question Answering Dataset for Benchmarking Vision-Language Models in Plant Science
PlantVillageVQA is a large-scale visual question answering (VQA) dataset derived from the widely used PlantVillage image corpus. It was designed to advance the development and evaluation of vision-language models for agricultural decision-making and analysis. The PlantVillageVQA dataset comprises 193,609 high-quality question-answer (QA) pairs grounded over 55,448 images spanning 14 crop species and 38 disease conditions. Questions are organised into 3 levels of cognitive complexity and 9 distinct categories. Each question category was phrased manually following expert guidance and generated via an automated two-stage pipeline: (1) template-based QA synthesis from image metadata and (2) multi-stage linguistic re-engineering. The dataset was iteratively reviewed by domain experts for scientific accuracy and relevancy. The final dataset was evaluated using three state-of-the-art models for quality assessment. Our objective remains to provide a publicly available, standardised and expert-verified database to enhance diagnostic accuracy for plant disease identifications and advance scientific research in the agricultural domain. Our dataset will be open-sourced at https://huggingface.co/datasets/SyedNazmusSakib/PlantVillageVQA.
comment: 17 pages, 15 figures and Submittd to Nature Scientific Data
SPIN-ODE: Stiff Physics-Informed Neural ODE for Chemical Reaction Rate Estimation ECAI
Estimating rate coefficients from complex chemical reactions is essential for advancing detailed chemistry. However, the stiffness inherent in real-world atmospheric chemistry systems poses severe challenges, leading to training instability and poor convergence, which hinder effective rate coefficient estimation using learning-based approaches. To address this, we propose a Stiff Physics-Informed Neural ODE framework (SPIN-ODE) for chemical reaction modelling. Our method introduces a three-stage optimisation process: first, a black-box neural ODE is trained to fit concentration trajectories; second, a Chemical Reaction Neural Network (CRNN) is pre-trained to learn the mapping between concentrations and their time derivatives; and third, the rate coefficients are fine-tuned by integrating with the pre-trained CRNN. Extensive experiments on both synthetic and newly proposed real-world datasets validate the effectiveness and robustness of our approach. As the first work addressing stiff neural ODE for chemical rate coefficient discovery, our study opens promising directions for integrating neural networks with detailed chemistry.
comment: Accepted at the European Conference on Artificial Intelligence (ECAI) 2025
FROG: Fair Removal on Graphs CIKM 2025
With growing emphasis on privacy regulations, machine unlearning has become increasingly critical in real-world applications such as social networks and recommender systems, many of which are naturally represented as graphs. However, existing graph unlearning methods often modify nodes or edges indiscriminately, overlooking their impact on fairness. For instance, forgetting links between users of different genders may inadvertently exacerbate group disparities. To address this issue, we propose a novel framework that jointly optimizes both the graph structure and the model to achieve fair unlearning. Our method rewires the graph by removing redundant edges that hinder forgetting while preserving fairness through targeted edge augmentation. We further introduce a worst-case evaluation mechanism to assess robustness under challenging scenarios. Experiments on real-world datasets show that our approach achieves more effective and fair unlearning than existing baselines.
comment: CIKM 2025
Cryptography and Security 31
Surveying the Operational Cybersecurity and Supply Chain Threat Landscape when Developing and Deploying AI Systems
The rise of AI has transformed the software and hardware landscape, enabling powerful capabilities through specialized infrastructures, large-scale data storage, and advanced hardware. However, these innovations introduce unique attack surfaces and objectives which traditional cybersecurity assessments often overlook. Cyber attackers are shifting their objectives from conventional goals like privilege escalation and network pivoting to manipulating AI outputs to achieve desired system effects, such as slowing system performance, flooding outputs with false positives, or degrading model accuracy. This paper serves to raise awareness of the novel cyber threats that are introduced when incorporating AI into a software system. We explore the operational cybersecurity and supply chain risks across the AI lifecycle, emphasizing the need for tailored security frameworks to address evolving threats in the AI-driven landscape. We highlight previous exploitations and provide insights from working in this area. By understanding these risks, organizations can better protect AI systems and ensure their reliability and resilience.
comment: 11 pages, 5 figures
Network-Level Prompt and Trait Leakage in Local Research Agents
We show that Web and Research Agents (WRAs) -- language model-based systems that investigate complex topics on the Internet -- are vulnerable to inference attacks by passive network adversaries such as ISPs. These agents could be deployed \emph{locally} by organizations and individuals for privacy, legal, or financial purposes. Unlike sporadic web browsing by humans, WRAs visit $70{-}140$ domains with distinguishable timing correlations, enabling unique fingerprinting attacks. Specifically, we demonstrate a novel prompt and user trait leakage attack against WRAs that only leverages their network-level metadata (i.e., visited IP addresses and their timings). We start by building a new dataset of WRA traces based on user search queries and queries generated by synthetic personas. We define a behavioral metric (called OBELS) to comprehensively assess similarity between original and inferred prompts, showing that our attack recovers over 73\% of the functional and domain knowledge of user prompts. Extending to a multi-session setting, we recover up to 19 of 32 latent traits with high accuracy. Our attack remains effective under partial observability and noisy conditions. Finally, we discuss mitigation strategies that constrain domain diversity or obfuscate traces, showing negligible utility impact while reducing attack effectiveness by an average of 29\%.
comment: under review
Robustness Assessment and Enhancement of Text Watermarking for Google's SynthID
Recent advances in LLM watermarking methods such as SynthID-Text by Google DeepMind offer promising solutions for tracing the provenance of AI-generated text. However, our robustness assessment reveals that SynthID-Text is vulnerable to meaning-preserving attacks, such as paraphrasing, copy-paste modifications, and back-translation, which can significantly degrade watermark detectability. To address these limitations, we propose SynGuard, a hybrid framework that combines the semantic alignment strength of Semantic Information Retrieval (SIR) with the probabilistic watermarking mechanism of SynthID-Text. Our approach jointly embeds watermarks at both lexical and semantic levels, enabling robust provenance tracking while preserving the original meaning. Experimental results across multiple attack scenarios show that SynGuard improves watermark recovery by an average of 11.1\% in F1 score compared to SynthID-Text. These findings demonstrate the effectiveness of semantic-aware watermarking in resisting real-world tampering. All code, datasets, and evaluation scripts are publicly available at: https://github.com/githshine/SynGuard.
comment: submitted to TrustCom2025
FlowMalTrans: Unsupervised Binary Code Translation for Malware Detection Using Flow-Adapter Architecture EMNLP 2025
Applying deep learning to malware detection has drawn great attention due to its notable performance. With the increasing prevalence of cyberattacks targeting IoT devices, there is a parallel rise in the development of malware across various Instruction Set Architectures (ISAs). It is thus important to extend malware detection capacity to multiple ISAs. However, training a deep learning-based malware detection model usually requires a large number of labeled malware samples. The process of collecting and labeling sufficient malware samples to build datasets for each ISA is labor-intensive and time-consuming. To reduce the burden of data collection, we propose to leverage the ideas of Neural Machine Translation (NMT) and Normalizing Flows (NFs) for malware detection. Specifically, when dealing with malware in a certain ISA, we translate it to an ISA with sufficient malware samples (like X86-64). This allows us to apply a model trained on one ISA to analyze malware from another ISA. Our approach reduces the data collection effort by enabling malware detection across multiple ISAs using a model trained on a single ISA.
comment: This paper is accepted to EMNLP 2025 Findings
AI Propaganda factories with language models
AI-powered influence operations can now be executed end-to-end on commodity hardware. We show that small language models produce coherent, persona-driven political messaging and can be evaluated automatically without human raters. Two behavioural findings emerge. First, persona-over-model: persona design explains behaviour more than model identity. Second, engagement as a stressor: when replies must counter-arguments, ideological adherence strengthens and the prevalence of extreme content increases. We demonstrate that fully automated influence-content production is within reach of both large and small actors. Consequently, defence should shift from restricting model access towards conversation-centric detection and disruption of campaigns and coordination infrastructure. Paradoxically, the very consistency that enables these operations also provides a detection signature.
Smart Contract Intent Detection with Pre-trained Programming Language Model
Malicious intent in smart contract development can lead to substantial economic losses. SmartIntentNN is a deep learning model specifically designed to identify unsafe intents in smart contracts. This model integrates the Universal Sentence Encoder, a K-means clustering-based intent highlighting mechanism, and a Bidirectional Long Short-Term Memory network for multi-label classification, achieving an F1 of 0.8633 in distinguishing ten different intent categories. In this study, we present an upgraded version of this model, SmartIntentNN2 (Smart Contract Intent Neural Network V2). A significant enhancement in V2 is the incorporation of a BERT-based pre-trained language model, which has been trained on a dataset of 16,000 real smart contracts using a Masked Language Modeling objective. SmartIntentNN2 retains the BiLSTM-based multi-label classification network. With an improved F1 of 0.927, V2 demonstrates enhanced performance compared to its predecessor, establishing itself as the state-of-the-art model for smart contract intent detection.
comment: 10 pages, 5 figures, conference
Disabling Self-Correction in Retrieval-Augmented Generation via Stealthy Retriever Poisoning
Retrieval-Augmented Generation (RAG) has become a standard approach for improving the reliability of large language models (LLMs). Prior work demonstrates the vulnerability of RAG systems by misleading them into generating attacker-chosen outputs through poisoning the knowledge base. However, this paper uncovers that such attacks could be mitigated by the strong \textit{self-correction ability (SCA)} of modern LLMs, which can reject false context once properly configured. This SCA poses a significant challenge for attackers aiming to manipulate RAG systems. In contrast to previous poisoning methods, which primarily target the knowledge base, we introduce \textsc{DisarmRAG}, a new poisoning paradigm that compromises the retriever itself to suppress the SCA and enforce attacker-chosen outputs. This compromisation enables the attacker to straightforwardly embed anti-SCA instructions into the context provided to the generator, thereby bypassing the SCA. To this end, we present a contrastive-learning-based model editing technique that performs localized and stealthy edits, ensuring the retriever returns a malicious instruction only for specific victim queries while preserving benign retrieval behavior. To further strengthen the attack, we design an iterative co-optimization framework that automatically discovers robust instructions capable of bypassing prompt-based defenses. We extensively evaluate DisarmRAG across six LLMs and three QA benchmarks. Our results show near-perfect retrieval of malicious instructions, which successfully suppress SCA and achieve attack success rates exceeding 90\% under diverse defensive prompts. Also, the edited retriever remains stealthy under several detection methods, highlighting the urgent need for retriever-centric defenses.
SCAMPER -- Synchrophasor Covert chAnnel for Malicious and Protective ERrands
We note that constituent fields (notably the fraction-of-seconds timestamp field) in the data payload structure of the synchrophasor communication protocol (IEEE C37.118 standard) are overprovisioned relative to real-world usage and needs, lending themselves to abuse for embedding of covert channels. We develop the SCAMPER (Synchrophasor Covert Channel for Malicious and Protective ERrands) framework to exploit these overprovisioned fields for covert communication and show that SCAMPER can be applied for both malicious (attack) and protective (defense) purposes. Through modifications of the timestamp field, we demonstrate that SCAMPER enables an attacker to accomplish surreptitious communications between devices in the power system to trigger a variety of malicious actions. These timestamp modifications can be performed without having any impact on the operation of the power system. However, having recognized the potential for this covert channel, we show that SCAMPER can instead be applied for defensive security purposes as an integrated cryptographic data integrity mechanism that can facilitate detection of false data injection (FDI) attacks. We perform experimental studies of the proposed methods on two Hardware-in-the-Loop (HIL) testbeds to demonstrate the effectiveness of the proposed SCAMPER framework for both malicious and protective purposes.
comment: 12 pages, 10 figures
SoK: Large Language Model Copyright Auditing via Fingerprinting
The broad capabilities and substantial resources required to train Large Language Models (LLMs) make them valuable intellectual property, yet they remain vulnerable to copyright infringement, such as unauthorized use and model theft. LLM fingerprinting, a non-intrusive technique that extracts and compares the distinctive features from LLMs to identify infringements, offers a promising solution to copyright auditing. However, its reliability remains uncertain due to the prevalence of diverse model modifications and the lack of standardized evaluation. In this SoK, we present the first comprehensive study of LLM fingerprinting. We introduce a unified framework and formal taxonomy that categorizes existing methods into white-box and black-box approaches, providing a structured overview of the state of the art. We further propose LeaFBench, the first systematic benchmark for evaluating LLM fingerprinting under realistic deployment scenarios. Built upon mainstream foundation models and comprising 149 distinct model instances, LeaFBench integrates 13 representative post-development techniques, spanning both parameter-altering methods (e.g., fine-tuning, quantization) and parameter-independent mechanisms (e.g., system prompts, RAG). Extensive experiments on LeaFBench reveal the strengths and weaknesses of existing methods, thereby outlining future research directions and critical open problems in this emerging field. The code is available at https://github.com/shaoshuo-ss/LeaFBench.
Every Keystroke You Make: A Tech-Law Measurement and Analysis of Event Listeners for Wiretapping
The privacy community has a long track record of investigating emerging types of web tracking techniques. Recent work has focused on compliance of web trackers with new privacy laws such as Europe's GDPR and California's CCPA. Despite the growing body of research documenting widespread lack of compliance with new privacy laws, there is a lack of robust enforcement. Different from prior work, we conduct a tech-law analysis to map decades-old U.S. laws about interception of electronic communications--so-called wiretapping--to web tracking. Bridging the tech-law gap for older wiretapping laws is important and timely because, in cases where legal harm to privacy is proven, they can provide statutory private right of action, are at the forefront of recent privacy enforcement, and could ultimately lead to a meaningful change in the web tracking landscape. In this paper, we focus on a particularly invasive tracking technique: the use of JavaScript event listeners by third-party trackers for real-time keystroke interception on websites. We use an instrumented web browser to crawl a sample of the top-million websites to investigate the use of event listeners that aligns with the criteria for wiretapping, according to U.S. wiretapping law at the federal level and in California. We find evidence that 38.52% websites installed third-party event listeners to intercept keystrokes, and that at least 3.18% websites transmitted intercepted information to a third-party server, which aligns with the criteria for wiretapping. We further find evidence that the intercepted information such as email addresses typed into form fields are used for unsolicited email marketing. Beyond our work that maps the intersection between technical measurement and U.S. wiretapping law, additional future legal research is required to determine when the wiretapping observed in our paper passes the threshold for illegality.
From Research to Reality: Feasibility of Gradient Inversion Attacks in Federated Learning KDD 2026
Gradient inversion attacks have garnered attention for their ability to compromise privacy in federated learning. However, many studies consider attacks with the model in inference mode, where training-time behaviors like dropout are disabled and batch normalization relies on fixed statistics. In this work, we systematically analyze how architecture and training behavior affect vulnerability, including the first in-depth study of inference-mode clients, which we show dramatically simplifies inversion. To assess attack feasibility under more realistic conditions, we turn to clients operating in standard training mode. In this setting, we find that successful attacks are only possible when several architectural conditions are met simultaneously: models must be shallow and wide, use skip connections, and, critically, employ pre-activation normalization. We introduce two novel attacks against models in training-mode with varying attacker knowledge, achieving state-of-the-art performance under realistic training conditions. We extend these efforts by presenting the first attack on a production-grade object-detection model. Here, to enable any visibly identifiable leakage, we revert to the lenient inference mode setting and make multiple architectural modifications to increase model vulnerability, with the extent of required changes highlighting the strong inherent robustness of such architectures. We conclude this work by offering the first comprehensive mapping of settings, clarifying which combinations of architectural choices and operational modes meaningfully impact privacy. Our analysis provides actionable insight into when models are likely vulnerable, when they appear robust, and where subtle leakage may persist. Together, these findings reframe how gradient inversion risk should be assessed in future research and deployment scenarios.
comment: Under review at KDD 2026 (Research Track)
The Art of Hide and Seek: Making Pickle-Based Model Supply Chain Poisoning Stealthy Again
Pickle deserialization vulnerabilities have persisted throughout Python's history, remaining widely recognized yet unresolved. Due to its ability to transparently save and restore complex objects into byte streams, many AI/ML frameworks continue to adopt pickle as the model serialization protocol despite its inherent risks. As the open-source model ecosystem grows, model-sharing platforms such as Hugging Face have attracted massive participation, significantly amplifying the real-world risks of pickle exploitation and opening new avenues for model supply chain poisoning. Although several state-of-the-art scanners have been developed to detect poisoned models, their incomplete understanding of the poisoning surface leaves the detection logic fragile and allows attackers to bypass them. In this work, we present the first systematic disclosure of the pickle-based model poisoning surface from both model loading and risky function perspectives. Our research demonstrates how pickle-based model poisoning can remain stealthy and highlights critical gaps in current scanning solutions. On the model loading surface, we identify 22 distinct pickle-based model loading paths across five foundational AI/ML frameworks, 19 of which are entirely missed by existing scanners. We further develop a bypass technique named Exception-Oriented Programming (EOP) and discover 9 EOP instances, 7 of which can bypass all scanners. On the risky function surface, we discover 133 exploitable gadgets, achieving almost a 100% bypass rate. Even against the best-performing scanner, these gadgets maintain an 89% bypass rate. By systematically revealing the pickle-based model poisoning surface, we achieve practical and robust bypasses against real-world scanners. We responsibly disclose our findings to corresponding vendors, receiving acknowledgments and a $6000 bug bounty.
Addressing Deepfake Issue in Selfie banking through camera based authentication
Fake images in selfie banking are increasingly becoming a threat. Previously, it was just Photoshop, but now deep learning technologies enable us to create highly realistic fake identities, which fraudsters exploit to bypass biometric systems such as facial recognition in online banking. This paper explores the use of an already established forensic recognition system, previously used for picture camera localization, in deepfake detection.
Safety Alignment Should Be Made More Than Just A Few Attention Heads
Current safety alignment for large language models(LLMs) continues to present vulnerabilities, given that adversarial prompting can effectively bypass their safety measures.Our investigation shows that these safety mechanisms predominantly depend on a limited subset of attention heads: removing or ablating these heads can severely compromise model safety. To identify and evaluate these safety-critical components, we introduce RDSHA, a targeted ablation method that leverages the model's refusal direction to pinpoint attention heads mostly responsible for safety behaviors. Further analysis shows that existing jailbreak attacks exploit this concentration by selectively bypassing or manipulating these critical attention heads. To address this issue, we propose AHD, a novel training strategy designed to promote the distributed encoding of safety-related behaviors across numerous attention heads. Experimental results demonstrate that AHD successfully distributes safety-related capabilities across more attention heads. Moreover, evaluations under several mainstream jailbreak attacks show that models trained with AHD exhibit considerably stronger safety robustness, while maintaining overall functional utility.
Intellectual Property in Graph-Based Machine Learning as a Service: Attacks and Defenses
Graph-structured data, which captures non-Euclidean relationships and interactions between entities, is growing in scale and complexity. As a result, training state-of-the-art graph machine learning (GML) models have become increasingly resource-intensive, turning these models and data into invaluable Intellectual Property (IP). To address the resource-intensive nature of model training, graph-based Machine-Learning-as-a-Service (GMLaaS) has emerged as an efficient solution by leveraging third-party cloud services for model development and management. However, deploying such models in GMLaaS also exposes them to potential threats from attackers. Specifically, while the APIs within a GMLaaS system provide interfaces for users to query the model and receive outputs, they also allow attackers to exploit and steal model functionalities or sensitive training data, posing severe threats to the safety of these GML models and the underlying graph data. To address these challenges, this survey systematically introduces the first taxonomy of threats and defenses at the level of both GML model and graph-structured data. Such a tailored taxonomy facilitates an in-depth understanding of GML IP protection. Furthermore, we present a systematic evaluation framework to assess the effectiveness of IP protection methods, introduce a curated set of benchmark datasets across various domains, and discuss their application scopes and future challenges. Finally, we establish an open-sourced versatile library named PyGIP, which evaluates various attack and defense techniques in GMLaaS scenarios and facilitates the implementation of existing benchmark methods. The library resource can be accessed at: https://labrai.github.io/PyGIP. We believe this survey will play a fundamental role in intellectual property protection for GML and provide practical recipes for the GML community.
A Scenario-Oriented Survey of Federated Recommender Systems: Techniques, Challenges, and Future Directions
Extending recommender systems to federated learning (FL) frameworks to protect the privacy of users or platforms while making recommendations has recently gained widespread attention in academia. This is due to the natural coupling of recommender systems and federated learning architectures: the data originates from distributed clients (mostly mobile devices held by users), which are highly related to privacy. In a centralized recommender system (CenRec), the central server collects clients' data, trains the model, and provides the service. Whereas in federated recommender systems (FedRec), the step of data collecting is omitted, and the step of model training is offloaded to each client. The server only aggregates the model and other knowledge, thus avoiding client privacy leakage. Some surveys of federated recommender systems discuss and analyze related work from the perspective of designing FL systems. However, their utility drops by ignoring specific recommendation scenarios' unique characteristics and practical challenges. For example, the statistical heterogeneity issue in cross-domain FedRec originates from the label drift of the data held by different platforms, which is mainly caused by the recommender itself, but not the federated architecture. Therefore, it should focus more on solving specific problems in real-world recommendation scenarios to encourage the deployment FedRec. To this end, this review comprehensively analyzes the coupling of recommender systems and federated learning from the perspective of recommendation researchers and practitioners. We establish a clear link between recommendation scenarios and FL frameworks, systematically analyzing scenario-specific approaches, practical challenges, and potential opportunities. We aim to develop guidance for the real-world deployment of FedRec, bridging the gap between existing research and applications.
Breaking the Layer Barrier: Remodeling Private Transformer Inference with Hybrid CKKS and MPC USENIX Security 2025
This paper presents an efficient framework for private Transformer inference that combines Homomorphic Encryption (HE) and Secure Multi-party Computation (MPC) to protect data privacy. Existing methods often leverage HE for linear layers (e.g., matrix multiplications) and MPC for non-linear layers (e.g., Softmax activation functions), but the conversion between HE and MPC introduces significant communication costs. The proposed framework, dubbed BLB, overcomes this by breaking down layers into fine-grained operators and further fusing adjacent linear operators, reducing the need for HE/MPC conversions. To manage the increased ciphertext bit width from the fused linear operators, BLB proposes the first secure conversion protocol between CKKS and MPC and enables CKKS-based computation of the fused operators. Additionally, BLB proposes an efficient matrix multiplication protocol for fused computation in Transformers. Extensive evaluations on BERT-base, BERT-large, and GPT2-base show that BLB achieves a $21\times$ reduction in communication overhead compared to BOLT (S\&P'24) and a $2\times$ reduction compared to Bumblebee (NDSS'25), along with latency reductions of $13\times$ and $1.8\times$, respectively, when leveraging GPU acceleration.
comment: USENIX Security 2025
Servant, Stalker, Predator: How An Honest, Helpful, And Harmless (3H) Agent Unlocks Adversarial Skills
This paper identifies and analyzes a novel vulnerability class in Model Context Protocol (MCP) based agent systems. The attack chain describes and demonstrates how benign, individually authorized tasks can be orchestrated to produce harmful emergent behaviors. Through systematic analysis using the MITRE ATLAS framework, we demonstrate how 95 agents tested with access to multiple services-including browser automation, financial analysis, location tracking, and code deployment-can chain legitimate operations into sophisticated attack sequences that extend beyond the security boundaries of any individual service. These red team exercises survey whether current MCP architectures lack cross-domain security measures necessary to detect or prevent a large category of compositional attacks. We present empirical evidence of specific attack chains that achieve targeted harm through service orchestration, including data exfiltration, financial manipulation, and infrastructure compromise. These findings reveal that the fundamental security assumption of service isolation fails when agents can coordinate actions across multiple domains, creating an exponential attack surface that grows with each additional capability. This research provides a barebones experimental framework that evaluate not whether agents can complete MCP benchmark tasks, but what happens when they complete them too well and optimize across multiple services in ways that violate human expectations and safety constraints. We propose three concrete experimental directions using the existing MCP benchmark suite.
Mind the Third Eye! Benchmarking Privacy Awareness in MLLM-powered Smartphone Agents
Smartphones bring significant convenience to users but also enable devices to extensively record various types of personal information. Existing smartphone agents powered by Multimodal Large Language Models (MLLMs) have achieved remarkable performance in automating different tasks. However, as the cost, these agents are granted substantial access to sensitive users' personal information during this operation. To gain a thorough understanding of the privacy awareness of these agents, we present the first large-scale benchmark encompassing 7,138 scenarios to the best of our knowledge. In addition, for privacy context in scenarios, we annotate its type (e.g., Account Credentials), sensitivity level, and location. We then carefully benchmark seven available mainstream smartphone agents. Our results demonstrate that almost all benchmarked agents show unsatisfying privacy awareness (RA), with performance remaining below 60% even with explicit hints. Overall, closed-source agents show better privacy ability than open-source ones, and Gemini 2.0-flash achieves the best, achieving an RA of 67%. We also find that the agents' privacy detection capability is highly related to scenario sensitivity level, i.e., the scenario with a higher sensitivity level is typically more identifiable. We hope the findings enlighten the research community to rethink the unbalanced utility-privacy tradeoff about smartphone agents. Our code and benchmark are available at https://zhixin-l.github.io/SAPA-Bench.
PoolFlip: A Multi-Agent Reinforcement Learning Security Environment for Cyber Defense
Cyber defense requires automating defensive decision-making under stealthy, deceptive, and continuously evolving adversarial strategies. The FlipIt game provides a foundational framework for modeling interactions between a defender and an advanced adversary that compromises a system without being immediately detected. In FlipIt, the attacker and defender compete to control a shared resource by performing a Flip action and paying a cost. However, the existing FlipIt frameworks rely on a small number of heuristics or specialized learning techniques, which can lead to brittleness and the inability to adapt to new attacks. To address these limitations, we introduce PoolFlip, a multi-agent gym environment that extends the FlipIt game to allow efficient learning for attackers and defenders. Furthermore, we propose Flip-PSRO, a multi-agent reinforcement learning (MARL) approach that leverages population-based training to train defender agents equipped to generalize against a range of unknown, potentially adaptive opponents. Our empirical results suggest that Flip-PSRO defenders are $2\times$ more effective than baselines to generalize to a heuristic attack not exposed in training. In addition, our newly designed ownership-based utility functions ensure that Flip-PSRO defenders maintain a high level of control while optimizing performance.
comment: Accepted at GameSec 2025
Quantum One-Time Programs, Revisited STOC 2025
One-time programs (Goldwasser, Kalai and Rothblum, CRYPTO 2008) are functions that can be run on any single input of a user's choice, but not on a second input. Classically, they are unachievable without trusted hardware, but the destructive nature of quantum measurements seems to provide a quantum path to constructing them. Unfortunately, Broadbent, Gutoski and Stebila showed that even with quantum techniques, a strong notion of one-time programs, similar to ideal obfuscation, cannot be achieved for any non-trivial quantum function. On the positive side, Ben-David and Sattath (Quantum, 2023) showed how to construct a one-time program for a certain (probabilistic) digital signature scheme, under a weaker notion of one-time program security. There is a vast gap between achievable and provably impossible notions of one-time program security, and it is unclear what functionalities are one-time programmable under the achievable notions of security. In this work, we present new, meaningful, yet achievable definitions of one-time program security for probabilistic classical functions. We show how to construct one time programs satisfying these definitions for all functions in the classical oracle model and for constrained pseudorandom functions in the plain model. Finally, we examine the limits of these notions: we show a class of functions which cannot be one-time programmed in the plain model, as well as a class of functions which appears to be highly random given a single query, but whose one-time program form leaks the entire function even in the oracle model.
comment: minor revision; in STOC 2025
When AIOps Become "AI Oops": Subverting LLM-driven IT Operations via Telemetry Manipulation
AI for IT Operations (AIOps) is transforming how organizations manage complex software systems by automating anomaly detection, incident diagnosis, and remediation. Modern AIOps solutions increasingly rely on autonomous LLM-based agents to interpret telemetry data and take corrective actions with minimal human intervention, promising faster response times and operational cost savings. In this work, we perform the first security analysis of AIOps solutions, showing that, once again, AI-driven automation comes with a profound security cost. We demonstrate that adversaries can manipulate system telemetry to mislead AIOps agents into taking actions that compromise the integrity of the infrastructure they manage. We introduce techniques to reliably inject telemetry data using error-inducing requests that influence agent behavior through a form of adversarial reward-hacking; plausible but incorrect system error interpretations that steer the agent's decision-making. Our attack methodology, AIOpsDoom, is fully automated--combining reconnaissance, fuzzing, and LLM-driven adversarial input generation--and operates without any prior knowledge of the target system. To counter this threat, we propose AIOpsShield, a defense mechanism that sanitizes telemetry data by exploiting its structured nature and the minimal role of user-generated content. Our experiments show that AIOpsShield reliably blocks telemetry-based attacks without affecting normal agent performance. Ultimately, this work exposes AIOps as an emerging attack vector for system compromise and underscores the urgent need for security-aware AIOps design.
comment: v0.2
DATABench: Evaluating Dataset Auditing in Deep Learning from an Adversarial Perspective
The widespread application of Deep Learning across diverse domains hinges critically on the quality and composition of training datasets. However, the common lack of disclosure regarding their usage raises significant privacy and copyright concerns. Dataset auditing techniques, which aim to determine if a specific dataset was used to train a given suspicious model, provide promising solutions to addressing these transparency gaps. While prior work has developed various auditing methods, their resilience against dedicated adversarial attacks remains largely unexplored. To bridge the gap, this paper initiates a comprehensive study evaluating dataset auditing from an adversarial perspective. We start with introducing a novel taxonomy, classifying existing methods based on their reliance on internal features (IF) (inherent to the data) versus external features (EF) (artificially introduced for auditing). Subsequently, we formulate two primary attack types: evasion attacks, designed to conceal the use of a dataset, and forgery attacks, intending to falsely implicate an unused dataset. Building on the understanding of existing methods and attack objectives, we further propose systematic attack strategies: decoupling, removal, and detection for evasion; adversarial example-based methods for forgery. These formulations and strategies lead to our new benchmark, DATABench, comprising 17 evasion attacks, 5 forgery attacks, and 9 representative auditing methods. Extensive evaluations using DATABench reveal that none of the evaluated auditing methods are sufficiently robust or distinctive under adversarial settings. These findings underscore the urgent need for developing a more secure and reliable dataset auditing method capable of withstanding sophisticated adversarial manipulation. Code is available at https://github.com/shaoshuo-ss/DATABench.
VideoEraser: Concept Erasure in Text-to-Video Diffusion Models EMNLP
The rapid growth of text-to-video (T2V) diffusion models has raised concerns about privacy, copyright, and safety due to their potential misuse in generating harmful or misleading content. These models are often trained on numerous datasets, including unauthorized personal identities, artistic creations, and harmful materials, which can lead to uncontrolled production and distribution of such content. To address this, we propose VideoEraser, a training-free framework that prevents T2V diffusion models from generating videos with undesirable concepts, even when explicitly prompted with those concepts. Designed as a plug-and-play module, VideoEraser can seamlessly integrate with representative T2V diffusion models via a two-stage process: Selective Prompt Embedding Adjustment (SPEA) and Adversarial-Resilient Noise Guidance (ARNG). We conduct extensive evaluations across four tasks, including object erasure, artistic style erasure, celebrity erasure, and explicit content erasure. Experimental results show that VideoEraser consistently outperforms prior methods regarding efficacy, integrity, fidelity, robustness, and generalizability. Notably, VideoEraser achieves state-of-the-art performance in suppressing undesirable content during T2V generation, reducing it by 46% on average across four tasks compared to baselines.
comment: To appear in the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP)
R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt Tuning CVPR 2025
Vision-language models (VLMs), such as CLIP, have gained significant popularity as foundation models, with numerous fine-tuning methods developed to enhance performance on downstream tasks. However, due to their inherent vulnerability and the common practice of selecting from a limited set of open-source models, VLMs suffer from a higher risk of adversarial attacks than traditional vision models. Existing defense techniques typically rely on adversarial fine-tuning during training, which requires labeled data and lacks of flexibility for downstream tasks. To address these limitations, we propose robust test-time prompt tuning (R-TPT), which mitigates the impact of adversarial attacks during the inference stage. We first reformulate the classic marginal entropy objective by eliminating the term that introduces conflicts under adversarial conditions, retaining only the pointwise entropy minimization. Furthermore, we introduce a plug-and-play reliability-based weighted ensembling strategy, which aggregates useful information from reliable augmented views to strengthen the defense. R-TPT enhances defense against adversarial attacks without requiring labeled training data while offering high flexibility for inference tasks. Extensive experiments on widely used benchmarks with various attacks demonstrate the effectiveness of R-TPT. The code is available in https://github.com/TomSheng21/R-TPT.
comment: CVPR 2025 (Corrected the results on the Aircraft dataset)
PromptKeeper: Safeguarding System Prompts for LLMs EMNLP 2025
System prompts are widely used to guide the outputs of large language models (LLMs). These prompts often contain business logic and sensitive information, making their protection essential. However, adversarial and even regular user queries can exploit LLM vulnerabilities to expose these hidden prompts. To address this issue, we propose PromptKeeper, a defense mechanism designed to safeguard system prompts by tackling two core challenges: reliably detecting leakage and mitigating side-channel vulnerabilities when leakage occurs. By framing detection as a hypothesis-testing problem, PromptKeeper effectively identifies both explicit and subtle leakage. Upon leakage detected, it regenerates responses using a dummy prompt, ensuring that outputs remain indistinguishable from typical interactions when no leakage is present. PromptKeeper ensures robust protection against prompt extraction attacks via either adversarial or regular queries, while preserving conversational capability and runtime efficiency during benign user interactions.
comment: Accepted to the Findings of EMNLP 2025. 17 pages, 6 figures, 3 tables
A Systematic Survey of Model Extraction Attacks and Defenses: State-of-the-Art and Perspectives
Machine learning (ML) models have significantly grown in complexity and utility, driving advances across multiple domains. However, substantial computational resources and specialized expertise have historically restricted their wide adoption. Machine-Learning-as-a-Service (MLaaS) platforms have addressed these barriers by providing scalable, convenient, and affordable access to sophisticated ML models through user-friendly APIs. While this accessibility promotes widespread use of advanced ML capabilities, it also introduces vulnerabilities exploited through Model Extraction Attacks (MEAs). Recent studies have demonstrated that adversaries can systematically replicate a target model's functionality by interacting with publicly exposed interfaces, posing threats to intellectual property, privacy, and system security. In this paper, we offer a comprehensive survey of MEAs and corresponding defense strategies. We propose a novel taxonomy that classifies MEAs according to attack mechanisms, defense approaches, and computing environments. Our analysis covers various attack techniques, evaluates their effectiveness, and highlights challenges faced by existing defenses, particularly the critical trade-off between preserving model utility and ensuring security. We further assess MEAs within different computing paradigms and discuss their technical, ethical, legal, and societal implications, along with promising directions for future research. This systematic survey aims to serve as a valuable reference for researchers, practitioners, and policymakers engaged in AI security and privacy. Additionally, we maintain an online repository continuously updated with related literature at https://github.com/kzhao5/ModelExtractionPapers.
Breaking the Gaussian Barrier: Residual-PAC Privacy for Automatic Privatization
The Probably Approximately Correct (PAC) Privacy framework [46] provides a powerful instance-based methodology to preserve privacy in complex data-driven systems. Existing PAC Privacy algorithms (we call them Auto-PAC) rely on a Gaussian mutual information upper bound. However, we show that the upper bound obtained by these algorithms is tight if and only if the perturbed mechanism output is jointly Gaussian with independent Gaussian noise. We propose two approaches for addressing this issue. First, we introduce two tractable post-processing methods for Auto-PAC, based on Donsker-Varadhan representation and sliced Wasserstein distances. However, the result still leaves wasted privacy budget. To address this issue more fundamentally, we introduce Residual-PAC (R-PAC) Privacy, an f-divergence-based measure to quantify privacy that remains after adversarial inference. To implement R-PAC Privacy in practice, we propose a Stackelberg Residual-PAC (SR-PAC) privatization mechanism, a game-theoretic framework that selects optimal noise distributions through convex bilevel optimization. Our approach achieves efficient privacy budget utilization for arbitrary data distributions and naturally composes when multiple mechanisms access the dataset. Through extensive experiments, we demonstrate that SR-PAC consistently obtains a better privacy-utility tradeoff than both PAC and differential privacy baselines.
MoveScanner: Analysis of Security Risks of Move Smart Contracts
As blockchain technology continues to evolve, the security of smart contracts has increasingly drawn attention from both academia and industry. The Move language, with its unique resource model and linear type system, provides a solid foundation for the security of digital assets. However, smart contracts still face new security challenges due to developer programming errors and the potential risks associated with cross-module interactions. This paper systematically analyzes the limitations of existing security tools within the Move ecosystem and reveals their unique vulnerability patterns. To address these issues, it introduces MoveScanner, a static analysis tool based on a control flow graph and data flow analysis architecture. By incorporating cross-module call graph tracking, MoveScanner can effectively identify five key types of security vulnerabilities, including resource leaks, weak permission management, and arithmetic overflows. In terms of design, MoveScanner adheres to a modular principle, supports bytecode-level analysis and multi-chain adaptation, and introduces innovative resource trajectory tracking algorithms and capability matrix analysis methods, thereby significantly reducing the false positive rate. Empirical results show that MoveScanner achieved 88.2% detection accuracy in benchmark testing, filling the gap in security tools in the Move ecosystem. Furthermore, this paper identifies twelve new types of security risks based on the resource-oriented programming paradigm and provides a theoretical foundation and practical experience for the development of smart contract security mechanisms. Future work will focus on combining formal verification and dynamic analysis techniques to build a security protection framework covering the entire contract lifecycle
Practical and Accurate Local Edge Differentially Private Graph Algorithms VLDB 2025
The rise of massive networks across diverse domains necessitates sophisticated graph analytics, often involving sensitive data and raising privacy concerns. This paper addresses these challenges using local differential privacy (LDP), which enforces privacy at the individual level, where no third-party entity is trusted, unlike centralized models that assume a trusted curator. We introduce novel LDP algorithms for two fundamental graph statistics: k-core decomposition and triangle counting. Our approach leverages input-dependent private graph properties, specifically the degeneracy and maximum degree of the graph, to improve theoretical utility. Unlike prior methods, our error bounds are determined by the maximum degree rather than the total number of edges, resulting in significantly tighter guarantees. For triangle counting, we improve upon the work of Imola, Murakami, and Chaudhury [USENIX Security `21, `22], which bounds error in terms of edge count. Instead, our algorithm achieves bounds based on graph degeneracy by leveraging a private out-degree orientation, a refined variant of Eden et al.'s randomized response technique [ICALP `23], and a novel analysis, yielding stronger guarantees than prior work. Beyond theoretical gains, we are the first to evaluate local DP algorithms in a distributed simulation, unlike prior work tested on a single processor. Experiments on real-world graphs show substantial accuracy gains: our k-core decomposition achieves errors within 3x of exact values, far outperforming the 131x error in the baseline of Dhulipala et al. [FOCS `22]. Our triangle counting algorithm reduces multiplicative approximation errors by up to six orders of magnitude, while maintaining competitive runtime.
comment: To appear in VLDB 2025
Revisiting Pre-trained Language Models for Vulnerability Detection
The rapid advancement of pre-trained language models (PLMs) has demonstrated promising results for various code-related tasks. However, their effectiveness in detecting real-world vulnerabilities remains a critical challenge. % for the security community. While existing empirical studies evaluate PLMs for vulnerability detection (VD), their inadequate consideration in data preparation, evaluation setups, and experimental settings undermines the accuracy and comprehensiveness of evaluations. This paper introduces RevisitVD, an extensive evaluation of 17 PLMs spanning smaller code-specific PLMs and large-scale PLMs using newly constructed datasets. Specifically, we compare the performance of PLMs under both fine-tuning and prompt engineering, assess their effectiveness and generalizability across various training and testing settings, and analyze their robustness against code normalization, abstraction, and semantic-preserving transformations. Our findings reveal that, for VD tasks, PLMs incorporating pre-training tasks designed to capture the syntactic and semantic patterns of code outperform both general-purpose PLMs and those solely pre-trained or fine-tuned on large code corpora. However, these models face notable challenges in real-world scenarios, such as difficulties in detecting vulnerabilities with complex dependencies, handling perturbations introduced by code normalization and abstraction, and identifying semantic-preserving vulnerable code transformations. Also, the truncation caused by the limited context windows of PLMs can lead to a non-negligible amount of labeling errors. This study underscores the importance of thorough evaluations of model performance in practical scenarios and outlines future directions to help enhance the effectiveness of PLMs for realistic VD applications.
Cryptography and Security 46
SIExVulTS: Sensitive Information Exposure Vulnerability Detection System using Transformer Models and Static Analysis
Sensitive Information Exposure (SIEx) vulnerabilities (CWE-200) remain a persistent and under-addressed threat across software systems, often leading to serious security breaches. Existing detection tools rarely target the diverse subcategories of CWE-200 or provide context-aware analysis of code-level data flows. Aims: This paper aims to present SIExVulTS, a novel vulnerability detection system that integrates transformer-based models with static analysis to identify and verify sensitive information exposure in Java applications. Method: SIExVulTS employs a three-stage architecture: (1) an Attack Surface Detection Engine that uses sentence embeddings to identify sensitive variables, strings, comments, and sinks; (2) an Exposure Analysis Engine that instantiates CodeQL queries aligned with the CWE-200 hierarchy; and (3) a Flow Verification Engine that leverages GraphCodeBERT to semantically validate source-to-sink flows. We evaluate SIExVulTS using three curated datasets, including real-world CVEs, a benchmark set of synthetic CWE-200 examples, and labeled flows from 31 open-source projects. Results: The Attack Surface Detection Engine achieved an average F1 score greater than 93\%, the Exposure Analysis Engine achieved an F1 score of 85.71\%, and the Flow Verification Engine increased precision from 22.61\% to 87.23\%. Moreover, SIExVulTS successfully uncovered six previously unknown CVEs in major Apache projects. Conclusions: The results demonstrate that SIExVulTS is effective and practical for improving software security against sensitive data exposure, addressing limitations of existing tools in detecting and verifying CWE-200 vulnerabilities.
Addressing Weak Authentication like RFID, NFC in EVs and EVCs using AI-powered Adaptive Authentication
The rapid expansion of the Electric Vehicles (EVs) and Electric Vehicle Charging Systems (EVCs) has introduced new cybersecurity challenges, specifically in authentication protocols that protect vehicles, users, and energy infrastructure. Although widely adopted for convenience, traditional authentication mechanisms like Radio Frequency Identification (RFID) and Near Field Communication (NFC) rely on static identifiers and weak encryption, making them highly vulnerable to attack vectors such as cloning, relay attacks, and signal interception. This study explores an AI-powered adaptive authentication framework designed to overcome these shortcomings by integrating machine learning, anomaly detection, behavioral analytics, and contextual risk assessment. Grounded in the principles of Zero Trust Architecture, the proposed framework emphasizes continuous verification, least privilege access, and secure communication. Through a comprehensive literature review, this research evaluates current vulnerabilities and highlights AI-driven solutions to provide a scalable, resilient, and proactive defense. Ultimately, the research findings conclude that adopting AI-powered adaptive authentication is a strategic imperative for securing the future of electric mobility and strengthening digital trust across the ecosystem. Keywords: weak authentication, RFID, NFC, ML, AI-powered adaptive authentication, relay attacks, cloning, eavesdropping, MITM attacks, Zero Trust Architecture
comment: Research paper exploring AI-driven adaptive authentication in the Electric Vehicle industry
Reliable Weak-to-Strong Monitoring of LLM Agents
We stress test monitoring systems for detecting covert misbehavior in autonomous LLM agents (e.g., secretly sharing private information). To this end, we systematize a monitor red teaming (MRT) workflow that incorporates: (1) varying levels of agent and monitor situational awareness; (2) distinct adversarial strategies to evade the monitor, such as prompt injection; and (3) two datasets and environments -- SHADE-Arena for tool-calling agents and our new CUA-SHADE-Arena, which extends TheAgentCompany, for computer-use agents. We run MRT on existing LLM monitor scaffoldings, which orchestrate LLMs and parse agent trajectories, alongside a new hybrid hierarchical-sequential scaffolding proposed in this work. Our empirical results yield three key findings. First, agent awareness dominates monitor awareness: an agent's knowledge that it is being monitored substantially degrades the monitor's reliability. On the contrary, providing the monitor with more information about the agent is less helpful than expected. Second, monitor scaffolding matters more than monitor awareness: the hybrid scaffolding consistently outperforms baseline monitor scaffolding, and can enable weaker models to reliably monitor stronger agents -- a weak-to-strong scaling effect. Third, in a human-in-the-loop setting where humans discuss with the LLM monitor to get an updated judgment for the agent's behavior, targeted human oversight is most effective; escalating only pre-flagged cases to human reviewers improved the TPR by approximately 15% at FPR = 0.01. Our work establishes a standard workflow for MRT, highlighting the lack of adversarial robustness for LLMs and humans when monitoring and detecting agent misbehavior. We release code, data, and logs to spur further research.
comment: 18 pages, 15 figures
The Sample Complexity of Membership Inference and Privacy Auditing
A membership-inference attack gets the output of a learning algorithm, and a target individual, and tries to determine whether this individual is a member of the training data or an independent sample from the same distribution. A successful membership-inference attack typically requires the attacker to have some knowledge about the distribution that the training data was sampled from, and this knowledge is often captured through a set of independent reference samples from that distribution. In this work we study how much information the attacker needs for membership inference by investigating the sample complexity-the minimum number of reference samples required-for a successful attack. We study this question in the fundamental setting of Gaussian mean estimation where the learning algorithm is given $n$ samples from a Gaussian distribution $\mathcal{N}(\mu,\Sigma)$ in $d$ dimensions, and tries to estimate $\hat\mu$ up to some error $\mathbb{E}[\|\hat \mu - \mu\|^2_{\Sigma}]\leq \rho^2 d$. Our result shows that for membership inference in this setting, $\Omega(n + n^2 \rho^2)$ samples can be necessary to carry out any attack that competes with a fully informed attacker. Our result is the first to show that the attacker sometimes needs many more samples than the training algorithm uses to train the model. This result has significant implications for practice, as all attacks used in practice have a restricted form that uses $O(n)$ samples and cannot benefit from $\omega(n)$ samples. Thus, these attacks may be underestimating the possibility of membership inference, and better attacks may be possible when information about the distribution is easy to obtain.
comment: 58 Pages
ReLATE+: Unified Framework for Adversarial Attack Detection, Classification, and Resilient Model Selection in Time-Series Classification
Minimizing computational overhead in time-series classification, particularly in deep learning models, presents a significant challenge due to the high complexity of model architectures and the large volume of sequential data that must be processed in real time. This challenge is further compounded by adversarial attacks, emphasizing the need for resilient methods that ensure robust performance and efficient model selection. To address this challenge, we propose ReLATE+, a comprehensive framework that detects and classifies adversarial attacks, adaptively selects deep learning models based on dataset-level similarity, and thus substantially reduces retraining costs relative to conventional methods that do not leverage prior knowledge, while maintaining strong performance. ReLATE+ first checks whether the incoming data is adversarial and, if so, classifies the attack type, using this insight to identify a similar dataset from a repository and enable the reuse of the best-performing associated model. This approach ensures strong performance while reducing the need for retraining, and it generalizes well across different domains with varying data distributions and feature spaces. Experiments show that ReLATE+ reduces computational overhead by an average of 77.68%, enhancing adversarial resilience and streamlining robust model selection, all without sacrificing performance, within 2.02% of Oracle.
comment: Under review at IEEE TSMC Journal. arXiv admin note: text overlap with arXiv:2503.07882
CITADEL: Continual Anomaly Detection for Enhanced Learning in IoT Intrusion Detection
The Internet of Things (IoT), with its high degree of interconnectivity and limited computational resources, is particularly vulnerable to a wide range of cyber threats. Intrusion detection systems (IDS) have been extensively studied to enhance IoT security, and machine learning-based IDS (ML-IDS) show considerable promise for detecting malicious activity. However, their effectiveness is often constrained by poor adaptability to emerging threats and the issue of catastrophic forgetting during continuous learning. To address these challenges, we propose CITADEL, a self-supervised continual learning framework designed to extract robust representations from benign data while preserving long-term knowledge through optimized memory consolidation mechanisms. CITADEL integrates a tabular-to-image transformation module, a memory-aware masked autoencoder for self-supervised representation learning, and a novelty detection component capable of identifying anomalies without dependence on labeled attack data. Our design enables the system to incrementally adapt to emerging behaviors while retaining its ability to detect previously observed threats. Experiments on multiple intrusion datasets demonstrate that CITADEL achieves up to a 72.9% improvement over the VAE-based lifelong anomaly detector (VLAD) in key detection and retention metrics, highlighting its effectiveness in dynamic IoT environments.
comment: Under review at IEEE IoTJ
A NIS2 pan-European registry for identifying and classifying essential and important entities
The NIS2 Directive establishes a common cybersecurity governance model across the European Union, requiring member states to identify, classify, and supervise essential and important entities. As part of a broader governance network, member states are also obligated to notify the European Commission, the Cooperation Group, and ENISA about their cybersecurity infrastructure landscape. This thesis presents an analysis of the NIS2 Directive in this context and translates its provisions into concrete technical requirements. These requirements inform the design and implementation of a modular, legally grounded registry system intended to support competent authorities across the EU in meeting their obligations. Using the Design Science Research methodology, the thesis transforms complex legal provisions into structured workflows, deterministic classification algorithms, and interactive dashboards. The resulting system automates key regulatory processes, including entity registration, classification, and notification, while enabling context-aware supervision and reducing administrative burden. It supports both automated and manual registration methods and introduces a contextual labeling system to handle edge cases, risk factors, and cross-directive dependencies. Although developed for the Norwegian regulatory ecosystem, the system is designed for adaptation by other member states with minimal modification. This thesis contributes a reusable framework that bridges legal interpretation and technical implementation, offering a scalable solution for national and EU-level NIS2 cybersecurity governance. It also identifies key limitations and outlines opportunities for future research and development.
Towards Quantum Machine Learning for Malicious Code Analysis
Classical machine learning (CML) has been extensively studied for malware classification. With the emergence of quantum computing, quantum machine learning (QML) presents a paradigm-shifting opportunity to improve malware detection, though its application in this domain remains largely unexplored. In this study, we investigate two hybrid quantum-classical models -- a Quantum Multilayer Perceptron (QMLP) and a Quantum Convolutional Neural Network (QCNN), for malware classification. Both models utilize angle embedding to encode malware features into quantum states. QMLP captures complex patterns through full qubit measurement and data re-uploading, while QCNN achieves faster training via quantum convolution and pooling layers that reduce active qubits. We evaluate both models on five widely used malware datasets -- API-Graph, EMBER-Domain, EMBER-Class, AZ-Domain, and AZ-Class, across binary and multiclass classification tasks. Our results show high accuracy for binary classification -- 95-96% on API-Graph, 91-92% on AZ-Domain, and 77% on EMBER-Domain. In multiclass settings, accuracy ranges from 91.6-95.7% on API-Graph, 41.7-93.6% on AZ-Class, and 60.7-88.1% on EMBER-Class. Overall, QMLP outperforms QCNN in complex multiclass tasks, while QCNN offers improved training efficiency at the cost of reduced accuracy.
comment: 6 pages, 3 figures, 2 tables. Accepted at the International Workshop on Quantum Computing and Reinforcement Learning (QCRL) @ IEEE Quantum Week 2025
Just Dork and Crawl: Measuring Illegal Online Gambling Defacement in Indonesian Websites
This study investigates the defacement of Indonesian websites by actors promoting illegal online gambling. Using a lightweight methodology that combines keyword-driven dorking with systematic crawling, we identified 453 defaced webpages within one month. Although dorking alone yielded a false positive rate of approximately 20.3\%, the integration of crawling and keyword-counting enabled reliable differentiation between true and false positives. Our measurements revealed diverse defacement behaviors, including repeat defacements (150 cases), fixed instances (129), keyword modifications (55), and redirections or hidden URL injections. In total, 8,837 unique third-party URLs spanning 5,930 domains were captured, with a small subset recurring across multiple sites. Website responses were inconsistent, with an average reaction time of 75.3 hours. These findings demonstrate that simple, reproducible techniques can provide meaningful insights into the scale, persistence, and dynamics of defacement, highlighting the importance of continuous measurement for strengthening defenses against online gambling activities.
comment: 6 pages, 2 figures, IEEE Conference
An Efficient Lightweight Blockchain for Decentralized IoT
The Internet of Things (IoT) is applied in various fields, and the number of physical devices connected to the IoT is increasingly growing. There are significant challenges to the IoT's growth and development, mainly due to the centralized nature and large-scale IoT networks. The emphasis on the decentralization of IoT's architecture can overcome challenges to IoT's capabilities. A promising decentralized platform for IoT is blockchain. Owing to IoT devices' limited resources, traditional consensus algorithms such as PoW and PoS in the blockchain are computationally expensive. Therefore, the PoA consensus algorithm is proposed in the blockchain consensus network for IoT. The PoA selects the validator as Turn-based selection (TBS) that needs optimization and faces system reliability, energy consumption, latency, and low scalability. We propose an efficient, lightweight blockchain for decentralizing IoT architecture by using virtualization and clustering to increase productivity and scalability to address these issues. We also introduce a novel PoA based on the Weight-Based-Selection (WBS) method for validators to validate transactions and add them to the blockchain. By simulation, we evaluated the performance of our proposed WBS method as opposed to TBS. The results show reduced energy consumption, and response time, and increased throughput.
SecureV2X: An Efficient and Privacy-Preserving System for Vehicle-to-Everything (V2X) Applications
Autonomous driving and V2X technologies have developed rapidly in the past decade, leading to improved safety and efficiency in modern transportation. These systems interact with extensive networks of vehicles, roadside infrastructure, and cloud resources to support their machine learning capabilities. However, the widespread use of machine learning in V2X systems raises issues over the privacy of the data involved. This is particularly concerning for smart-transit and driver safety applications which can implicitly reveal user locations or explicitly disclose medical data such as EEG signals. To resolve these issues, we propose SecureV2X, a scalable, multi-agent system for secure neural network inferences deployed between the server and each vehicle. Under this setting, we study two multi-agent V2X applications: secure drowsiness detection, and secure red-light violation detection. Our system achieves strong performance relative to baselines, and scales efficiently to support a large number of secure computation interactions simultaneously. For instance, SecureV2X is $9.4 \times$ faster, requires $143\times$ fewer computational rounds, and involves $16.6\times$ less communication on drowsiness detection compared to other secure systems. Moreover, it achieves a runtime nearly $100\times$ faster than state-of-the-art benchmarks in object detection tasks for red light violation detection.
comment: 10 pages, 3 figures
Deep Data Hiding for ICAO-Compliant Face Images: A Survey
ICAO-compliant facial images, initially designed for secure biometric passports, are increasingly becoming central to identity verification in a wide range of application contexts, including border control, digital travel credentials, and financial services. While their standardization enables global interoperability, it also facilitates practices such as morphing and deepfakes, which can be exploited for harmful purposes like identity theft and illegal sharing of identity documents. Traditional countermeasures like Presentation Attack Detection (PAD) are limited to real-time capture and offer no post-capture protection. This survey paper investigates digital watermarking and steganography as complementary solutions that embed tamper-evident signals directly into the image, enabling persistent verification without compromising ICAO compliance. We provide the first comprehensive analysis of state-of-the-art techniques to evaluate the potential and drawbacks of the underlying approaches concerning the applications involving ICAO-compliant images and their suitability under standard constraints. We highlight key trade-offs, offering guidance for secure deployment in real-world identity systems.
comment: In 2025 IEEE International Joint Conference on Biometrics (IJCB)
A Technical Review on Comparison and Estimation of Steganographic Tools
Steganography is technique of hiding a data under cover media using different steganography tools. Image steganography is hiding of data (Text/Image/Audio/Video) under a cover as Image. This review paper presents classification of image steganography and the comparison of various Image steganography tools using different image formats. Analyzing numerous tools on the basis of Image features and extracting the best one. Some of the tools available in the market were selected based on the frequent use; these tools were tested using the same input on all of them. Specific text was embedded within all host images for each of the six Steganography tools selected. The results of the experiment reveal that all the six tools were relatively performing at the same level, though some software performs better than others through efficiency. And it was based on the image features like size, dimensions, and pixel value and histogram differentiation.
comment: 20
An Investigation on Group Query Hallucination Attacks
With the widespread use of large language models (LLMs), understanding their potential failure modes during user interactions is essential. In practice, users often pose multiple questions in a single conversation with LLMs. Therefore, in this study, we propose Group Query Attack, a technique that simulates this scenario by presenting groups of queries to LLMs simultaneously. We investigate how the accumulated context from consecutive prompts influences the outputs of LLMs. Specifically, we observe that Group Query Attack significantly degrades the performance of models fine-tuned on specific tasks. Moreover, we demonstrate that Group Query Attack induces a risk of triggering potential backdoors of LLMs. Besides, Group Query Attack is also effective in tasks involving reasoning, such as mathematical reasoning and code generation for pre-trained and aligned models.
Attackers Strike Back? Not Anymore -- An Ensemble of RL Defenders Awakens for APT Detection
Advanced Persistent Threats (APTs) represent a growing menace to modern digital infrastructure. Unlike traditional cyberattacks, APTs are stealthy, adaptive, and long-lasting, often bypassing signature-based detection systems. This paper introduces a novel framework for APT detection that unites deep learning, reinforcement learning (RL), and active learning into a cohesive, adaptive defense system. Our system combines auto-encoders for latent behavioral encoding with a multi-agent ensemble of RL-based defenders, each trained to distinguish between benign and malicious process behaviors. We identify a critical challenge in existing detection systems: their static nature and inability to adapt to evolving attack strategies. To this end, our architecture includes multiple RL agents (Q-Learning, PPO, DQN, adversarial defenders), each analyzing latent vectors generated by an auto-encoder. When any agent is uncertain about its decision, the system triggers an active learning loop to simulate expert feedback, thus refining decision boundaries. An ensemble voting mechanism, weighted by each agent's performance, ensures robust final predictions.
mmKey: Channel-Aware Beam Shaping for Reliable Key Generation in mmWave Wireless Networks
Physical-layer key generation (PLKG) has emerged as a promising technique to secure next-generation wireless networks by exploiting the inherent properties of the wireless channel. However, PLKG faces fundamental challenges in the millimeter wave (mmWave) regime due to channel sparsity, higher phase noise, and higher path loss, which undermine both the randomness and reciprocity required for secure key generation. In this paper, we present mmKey, a novel PLKG framework that capitalizes on the availability of multiple antennas at mmWave wireless nodes to inject randomness into an otherwise quasi-static wireless channel. Different from prior works that sacrifice either the secrecy of the key generation or the robustness, mmKey balances these two requirements. In particular, mmKey leverages a genetic algorithm to gradually evolve the initial weight vector population toward configurations that suppress the LOS component while taking into account the channel conditions, specifically, the sparsity and the signal-to-noise ratio (SNR). Extensive simulations show that mmKey improves the secrecy gap by an average of 39.4% over random beamforming and 34.0% over null beamforming, outperforming conventional schemes.
The Double-edged Sword of LLM-based Data Reconstruction: Understanding and Mitigating Contextual Vulnerability in Word-level Differential Privacy Text Sanitization CCS 2025
Differentially private text sanitization refers to the process of privatizing texts under the framework of Differential Privacy (DP), providing provable privacy guarantees while also empirically defending against adversaries seeking to harm privacy. Despite their simplicity, DP text sanitization methods operating at the word level exhibit a number of shortcomings, among them the tendency to leave contextual clues from the original texts due to randomization during sanitization $\unicode{x2013}$ this we refer to as $\textit{contextual vulnerability}$. Given the powerful contextual understanding and inference capabilities of Large Language Models (LLMs), we explore to what extent LLMs can be leveraged to exploit the contextual vulnerability of DP-sanitized texts. We expand on previous work not only in the use of advanced LLMs, but also in testing a broader range of sanitization mechanisms at various privacy levels. Our experiments uncover a double-edged sword effect of LLM-based data reconstruction attacks on privacy and utility: while LLMs can indeed infer original semantics and sometimes degrade empirical privacy protections, they can also be used for good, to improve the quality and privacy of DP-sanitized texts. Based on our findings, we propose recommendations for using LLM data reconstruction as a post-processing step, serving to increase privacy protection by thinking adversarially.
comment: 15 pages, 4 figures, 8 tables. Accepted to WPES @ CCS 2025
LLMs in the SOC: An Empirical Study of Human-AI Collaboration in Security Operations Centres
The integration of Large Language Models (LLMs) into Security Operations Centres (SOCs) presents a transformative, yet still evolving, opportunity to reduce analyst workload through human-AI collaboration. However, their real-world application in SOCs remains underexplored. To address this gap, we present a longitudinal study of 3,090 analyst queries from 45 SOC analysts over 10 months. Our analysis reveals that analysts use LLMs as on-demand aids for sensemaking and context-building, rather than for making high-stakes determinations, preserving analyst decision authority. The majority of queries are related to interpreting low-level telemetry (e.g., commands) and refining technical communication through short (1-3 turn) interactions. Notably, 93% of queries align with established cybersecurity competencies (NICE Framework), underscoring the relevance of LLM use for SOC-related tasks. Despite variations in tasks and engagement, usage trends indicate a shift from occasional exploration to routine integration, with growing adoption and sustained use among a subset of analysts. We find that LLMs function as flexible, on-demand cognitive aids that augment, rather than replace, SOC expertise. Our study provides actionable guidance for designing context-aware, human-centred AI assistance in security operations, highlighting the need for further in-the-wild research on real-world analyst-LLM collaboration, challenges, and impacts.
comment: 22 pages, 9 figures, under review
EnerSwap: Large-Scale, Privacy-First Automated Market Maker for V2G Energy Trading
With the rapid growth of Electric Vehicle (EV) technology, EVs are destined to shape the future of transportation. The large number of EVs facilitates the development of the emerging vehicle-to-grid (V2G) technology, which realizes bidirectional energy exchanges between EVs and the power grid. This has led to the setting up of electricity markets that are usually confined to a small geographical location, often with a small number of participants. Usually, these markets are manipulated by intermediaries responsible for collecting bids from prosumers, determining the market-clearing price, incorporating grid constraints, and accounting for network losses. While centralized models can be highly efficient, they grant excessive power to the intermediary by allowing them to gain exclusive access to prosumers \textquotesingle price preferences. This opens the door to potential market manipulation and raises significant privacy concerns for users, such as the location of energy providers. This lack of protection exposes users to potential risks, as untrustworthy servers and malicious adversaries can exploit this information to infer trading activities and real identities. This work proposes a secure, decentralized exchange market built on blockchain technology, utilizing a privacy-preserving Automated Market Maker (AMM) model to offer open and fair, and equal access to traders, and mitigates the most common trading-manipulation attacks. Additionally, it incorporates a scalable architecture based on geographical dynamic sharding, allowing for efficient resource allocation and improved performance as the market grows.
comment: 11 pages, 7 figures, 1 table, 1 algorithm, Paper accepted in 27th MSWiM Conference
DRMD: Deep Reinforcement Learning for Malware Detection under Concept Drift
Malware detection in real-world settings must deal with evolving threats, limited labeling budgets, and uncertain predictions. Traditional classifiers, without additional mechanisms, struggle to maintain performance under concept drift in malware domains, as their supervised learning formulation cannot optimize when to defer decisions to manual labeling and adaptation. Modern malware detection pipelines combine classifiers with monthly active learning (AL) and rejection mechanisms to mitigate the impact of concept drift. In this work, we develop a novel formulation of malware detection as a one-step Markov Decision Process and train a deep reinforcement learning (DRL) agent, simultaneously optimizing sample classification performance and rejecting high-risk samples for manual labeling. We evaluated the joint detection and drift mitigation policy learned by the DRL-based Malware Detection (DRMD) agent through time-aware evaluations on Android malware datasets subject to realistic drift requiring multi-year performance stability. The policies learned under these conditions achieve a higher Area Under Time (AUT) performance compared to standard classification approaches used in the domain, showing improved resilience to concept drift. Specifically, the DRMD agent achieved a $5.18\pm5.44$, $14.49\pm12.86$, and $10.06\pm10.81$ average AUT performance improvement for the classification only, classification with rejection, and classification with rejection and AL settings, respectively. Our results demonstrate for the first time that DRL can facilitate effective malware detection and improved resiliency to concept drift in the dynamic environment of the Android malware domain.
comment: 10 pages
A Tight Context-aware Privacy Bound for Histogram Publication
We analyze the privacy guarantees of the Laplace mechanism releasing the histogram of a dataset through the lens of pointwise maximal leakage (PML). While differential privacy is commonly used to quantify the privacy loss, it is a context-free definition that does not depend on the data distribution. In contrast, PML enables a more refined analysis by incorporating assumptions about the data distribution. We show that when the probability of each histogram bin is bounded away from zero, stronger privacy protection can be achieved for a fixed level of noise. Our results demonstrate the advantage of context-aware privacy measures and show that incorporating assumptions about the data can improve privacy-utility tradeoffs.
comment: Submitted to IEEE Signal Processing Letters
Quantum computing on encrypted data with arbitrary rotation gates
An efficient technique of computing on encrypted data allows a client with limited capability to perform complex operations on a remote fault-tolerant server without leaking anything about the input or output. Quantum computing provides information-theoretic security to solve such a problem, and many such techniques have been proposed under the premises of half-blind quantum computation. However, they are dependent on a fixed non-parametric resource set that comprises some universal combination of $H,S,T,CX, CZ$ or $CCX$ gates. In this study, we show that recursive decryption of the parametric gate, $R_z(\theta)$, is possible exactly when $\theta=\pm\pi/2^m$ for $m\in \mathbb{Z^{+}}$, and approximately with arbitrary precision $\epsilon$ for given $\theta$. We also show that a blind algorithm based on such a technique needs at most $O(\log_2^2(\pi/\epsilon))$ computation steps and communication rounds, while the techniques based on a non-parametric resource set require $O(\ln^{3.97}(1/\epsilon))$ rounds. We use these results to propose a universal scheme of half-blind quantum computation for computing on encrypted data using arbitrary rotation gates. This substantial reduction in the depth of blind circuit is an affirmative step towards the practical application of such techniques in secure NISQ-era computing.
Hidden Tail: Adversarial Image Causing Stealthy Resource Consumption in Vision-Language Models
Vision-Language Models (VLMs) are increasingly deployed in real-world applications, but their high inference cost makes them vulnerable to resource consumption attacks. Prior attacks attempt to extend VLM output sequences by optimizing adversarial images, thereby increasing inference costs. However, these extended outputs often introduce irrelevant abnormal content, compromising attack stealthiness. This trade-off between effectiveness and stealthiness poses a major limitation for existing attacks. To address this challenge, we propose \textit{Hidden Tail}, a stealthy resource consumption attack that crafts prompt-agnostic adversarial images, inducing VLMs to generate maximum-length outputs by appending special tokens invisible to users. Our method employs a composite loss function that balances semantic preservation, repetitive special token induction, and suppression of the end-of-sequence (EOS) token, optimized via a dynamic weighting strategy. Extensive experiments show that \textit{Hidden Tail} outperforms existing attacks, increasing output length by up to 19.2$\times$ and reaching the maximum token limit, while preserving attack stealthiness. These results highlight the urgent need to improve the robustness of VLMs against efficiency-oriented adversarial threats. Our code is available at https://github.com/zhangrui4041/Hidden_Tail.
Leveraging 3D Technologies for Hardware Security: Opportunities and Challenges
3D die stacking and 2.5D interposer design are promising technologies to improve integration density, performance and cost. Current approaches face serious issues in dealing with emerging security challenges such as side channel attacks, hardware trojans, secure IC manufacturing and IP piracy. By utilizing intrinsic characteristics of 2.5D and 3D technologies, we propose novel opportunities in designing secure systems. We present: (i) a 3D architecture for shielding side-channel information; (ii) split fabrication using active interposers; (iii) circuit camouflage on monolithic 3D IC, and (iv) 3D IC-based security processing-in-memory (PIM). Advantages and challenges of these designs are discussed, showing that the new designs can improve existing countermeasures against security threats and further provide new security features.
Immutable Digital Recognition via Blockchain
The process integrates the decentralised management and centralised operation models, aligning them with the national policy directives. The developed solution enables the full utilisation of blockchain technology's advantages while also fostering community participation. Consequently, it establishes a secure, legal, reliable, and dynamic electronic certification system.
FALCON: Autonomous Cyber Threat Intelligence Mining with LLMs for IDS Rule Generation
Signature-based Intrusion Detection Systems (IDS) detect malicious activities by matching network or host activity against predefined rules. These rules are derived from extensive Cyber Threat Intelligence (CTI), which includes attack signatures and behavioral patterns obtained through automated tools and manual threat analysis, such as sandboxing. The CTI is then transformed into actionable rules for the IDS engine, enabling real-time detection and prevention. However, the constant evolution of cyber threats necessitates frequent rule updates, which delay deployment time and weaken overall security readiness. Recent advancements in agentic systems powered by Large Language Models (LLMs) offer the potential for autonomous IDS rule generation with internal evaluation. We introduce FALCON, an autonomous agentic framework that generates deployable IDS rules from CTI data in real-time and evaluates them using built-in multi-phased validators. To demonstrate versatility, we target both network (Snort) and host-based (YARA) mediums and construct a comprehensive dataset of IDS rules with their corresponding CTIs. Our evaluations indicate FALCON excels in automatic rule generation, with an average of 95% accuracy validated by qualitative evaluation with 84% inter-rater agreement among multiple cybersecurity analysts across all metrics. These results underscore the feasibility and effectiveness of LLM-driven data mining for real-time cyber threat mitigation.
comment: 11 pages, 5 figures, 4 tables
Membership Inference Attacks on LLM-based Recommender Systems
Large language models (LLMs) based Recommender Systems (RecSys) can flexibly adapt recommendation systems to different domains. It utilizes in-context learning (ICL), i.e., the prompts, to customize the recommendation functions, which include sensitive historical user-specific item interactions, e.g., implicit feedback like clicked items or explicit product reviews. Such private information may be exposed to novel privacy attack. However, no study has been done on this important issue. We design four membership inference attacks (MIAs), aiming to reveal whether victims' historical interactions have been used by system prompts. They are \emph{direct inquiry, hallucination, similarity, and poisoning attacks}, each of which utilizes the unique features of LLMs or RecSys. We have carefully evaluated them on three LLMs that have been used to develop ICL-LLM RecSys and two well-known RecSys benchmark datasets. The results confirm that the MIA threat on LLM RecSys is realistic: direct inquiry and poisoning attacks showing significantly high attack advantages. We have also analyzed the factors affecting these attacks, such as the number of shots in system prompts and the position of the victim in the shots.
UniC-RAG: Universal Knowledge Corruption Attacks to Retrieval-Augmented Generation
Retrieval-augmented generation (RAG) systems are widely deployed in real-world applications in diverse domains such as finance, healthcare, and cybersecurity. However, many studies showed that they are vulnerable to knowledge corruption attacks, where an attacker can inject adversarial texts into the knowledge database of a RAG system to induce the LLM to generate attacker-desired outputs. Existing studies mainly focus on attacking specific queries or queries with similar topics (or keywords). In this work, we propose UniC-RAG, a universal knowledge corruption attack against RAG systems. Unlike prior work, UniC-RAG jointly optimizes a small number of adversarial texts that can simultaneously attack a large number of user queries with diverse topics and domains, enabling an attacker to achieve various malicious objectives, such as directing users to malicious websites, triggering harmful command execution, or launching denial-of-service attacks. We formulate UniC-RAG as an optimization problem and further design an effective solution to solve it, including a balanced similarity-based clustering method to enhance the attack's effectiveness. Our extensive evaluations demonstrate that UniC-RAG is highly effective and significantly outperforms baselines. For instance, UniC-RAG could achieve over 90% attack success rate by injecting 100 adversarial texts into a knowledge database with millions of texts to simultaneously attack a large set of user queries (e.g., 2,000). Additionally, we evaluate existing defenses and show that they are insufficient to defend against UniC-RAG, highlighting the need for new defense mechanisms in RAG systems.
comment: 21 pages, 4 figures
PRISM: Robust VLM Alignment with Principled Reasoning for Integrated Safety in Multimodality
Safeguarding vision-language models (VLMs) is a critical challenge, as existing methods often suffer from over-defense, which harms utility, or rely on shallow alignment, failing to detect complex threats that require deep reasoning. To this end, we introduce PRISM (Principled Reasoning for Integrated Safety in Multimodality), a system2-like framework that aligns VLMs by embedding a structured, safety-aware reasoning process. Our framework consists of two key components: PRISM-CoT, a dataset that teaches safety-aware chain-of-thought reasoning, and PRISM-DPO, generated via Monte Carlo Tree Search (MCTS) to further refine this reasoning through Direct Preference Optimization to help obtain a delicate safety boundary. Comprehensive evaluations demonstrate PRISM's effectiveness, achieving remarkably low attack success rates including 0.15% on JailbreakV-28K for Qwen2-VL and 90% improvement over the previous best method on VLBreak for LLaVA-1.5. PRISM also exhibits strong robustness against adaptive attacks, significantly increasing computational costs for adversaries, and generalizes effectively to out-of-distribution challenges, reducing attack success rates to just 8.70% on the challenging multi-image MIS benchmark. Remarkably, this robust defense is achieved while preserving, and in some cases enhancing, model utility. To promote reproducibility, we have made our code, data, and model weights available at https://github.com/SaFoLab-WISC/PRISM.
MEraser: An Effective Fingerprint Erasure Approach for Large Language Models ACL 2025
Large Language Models (LLMs) have become increasingly prevalent across various sectors, raising critical concerns about model ownership and intellectual property protection. Although backdoor-based fingerprinting has emerged as a promising solution for model authentication, effective attacks for removing these fingerprints remain largely unexplored. Therefore, we present Mismatched Eraser (MEraser), a novel method for effectively removing backdoor-based fingerprints from LLMs while maintaining model performance. Our approach leverages a two-phase fine-tuning strategy utilizing carefully constructed mismatched and clean datasets. Through extensive evaluation across multiple LLM architectures and fingerprinting methods, we demonstrate that MEraser achieves complete fingerprinting removal while maintaining model performance with minimal training data of fewer than 1,000 samples. Furthermore, we introduce a transferable erasure mechanism that enables effective fingerprinting removal across different models without repeated training. In conclusion, our approach provides a practical solution for fingerprinting removal in LLMs, reveals critical vulnerabilities in current fingerprinting techniques, and establishes comprehensive evaluation benchmarks for developing more resilient model protection methods in the future.
comment: Accepted by ACL 2025, Main Conference, Long Paper
A Statistical Framework of Watermarks for Large Language Models: Pivot, Detection Efficiency and Optimal Rules
Since ChatGPT was introduced in November 2022, embedding (nearly) unnoticeable statistical signals into text generated by large language models (LLMs), also known as watermarking, has been used as a principled approach to provable detection of LLM-generated text from its human-written counterpart. In this paper, we introduce a general and flexible framework for reasoning about the statistical efficiency of watermarks and designing powerful detection rules. Inspired by the hypothesis testing formulation of watermark detection, our framework starts by selecting a pivotal statistic of the text and a secret key -- provided by the LLM to the verifier -- to enable controlling the false positive rate (the error of mistakenly detecting human-written text as LLM-generated). Next, this framework allows one to evaluate the power of watermark detection rules by obtaining a closed-form expression of the asymptotic false negative rate (the error of incorrectly classifying LLM-generated text as human-written). Our framework further reduces the problem of determining the optimal detection rule to solving a minimax optimization program. We apply this framework to two representative watermarks -- one of which has been internally implemented at OpenAI -- and obtain several findings that can be instrumental in guiding the practice of implementing watermarks. In particular, we derive optimal detection rules for these watermarks under our framework. These theoretically derived detection rules are demonstrated to be competitive and sometimes enjoy a higher power than existing detection approaches through numerical experiments.
comment: Accepted by Annals of Statistics
Anonymity-washing
Anonymization is a foundational principle of data privacy regulation, yet its practical application remains riddled with ambiguity and inconsistency. This paper introduces the concept of anonymity-washing -- the misrepresentation of the anonymity level of ``sanitized'' personal data -- as a critical privacy concern. While both legal and technical critiques of anonymization exist, they tend to address isolated aspects of the problem. In contrast, this paper offers a comprehensive overview of the conditions that enable anonymity-washing. It synthesizes fragmented legal interpretations, technical misunderstandings, and outdated regulatory guidance and complements them with a systematic review of national and international resources, including legal cases, data protection authority guidelines, and technical documentation. Our findings reveal a lack of coherent support for practitioners, contributing to the persistent misuse of pseudonymization and obsolete anonymization techniques. We conclude by recommending targeted education, clearer technical guidance, and closer cooperation between regulators, researchers, and industry to bridge the gap between legal norms and technical reality.
comment: Accepted at Privacy Forum
Generative Artificial Intelligence-Supported Pentesting: A Comparison between Claude Opus, GPT-4, and Copilot
The advent of Generative Artificial Intelligence (GenAI) has brought a significant change to our society. GenAI can be applied across numerous fields, with particular relevance in cybersecurity. Among the various areas of application, its use in penetration testing (pentesting) or ethical hacking processes is of special interest. In this paper, we have analyzed the potential of leading generic-purpose GenAI tools-Claude Opus, GPT-4 from ChatGPT, and Copilot-in augmenting the penetration testing process as defined by the Penetration Testing Execution Standard (PTES). Our analysis involved evaluating each tool across all PTES phases within a controlled virtualized environment. The findings reveal that, while these tools cannot fully automate the pentesting process, they provide substantial support by enhancing efficiency and effectiveness in specific tasks. Notably, all tools demonstrated utility; however, Claude Opus consistently outperformed the others in our experimental scenarios.
$AutoGuardX$: A Comprehensive Cybersecurity Framework for Connected Vehicles
The rapid integration of Internet of Things (IoT) and interconnected systems in modern vehicles not only introduced a new era of convenience, automation, and connected vehicles but also elevated their exposure to sophisticated cyber threats. This is especially evident in US and Canada, where cyber-enabled auto theft has surged in recent years, revealing the limitations of existing security measures for connected vehicles. In response, this paper proposes $AutoGuardX$, a comprehensive cybersecurity framework designed specifically for connected vehicles. $AutoGuardX$ combines key elements from existing recognized standards for vehicle security, such as ISO/SAE 21434 and ISO 26262, with advanced technologies, including machine learning-based anomaly detection, IoT security protocols, and encrypted communication channels. The framework addresses major attack vectors like relay attacks, controller area network (CAN) bus intrusions, and vulnerabilities introduced by emerging technologies such as 5G and quantum computing. $AutoGuardX$ is extensively evaluated through security simulations across a mix of Sedans and SUVs from four major vehicle brands manufactured between 2019 and 2023. The results demonstrate the framework's adaptability, scalability, and practical effectiveness against existing and emerging threats.
comment: 16 pages, 3 figures, 8 tables
Seal Your Backdoor with Variational Defense ICCV 2025
We propose VIBE, a model-agnostic framework that trains classifiers resilient to backdoor attacks. The key concept behind our approach is to treat malicious inputs and corrupted labels from the training dataset as observed random variables, while the actual clean labels are latent. VIBE then recovers the corresponding latent clean label posterior through variational inference. The resulting training procedure follows the expectation-maximization (EM) algorithm. The E-step infers the clean pseudolabels by solving an entropy-regularized optimal transport problem, while the M-step updates the classifier parameters via gradient descent. Being modular, VIBE can seamlessly integrate with recent advancements in self-supervised representation learning, which enhance its ability to resist backdoor attacks. We experimentally validate the method effectiveness against contemporary backdoor attacks on standard datasets, a large-scale setup with 1$k$ classes, and a dataset poisoned with multiple attacks. VIBE consistently outperforms previous defenses across all tested scenarios.
comment: Accepted to ICCV 2025
A Survey of Threats Against Voice Authentication and Anti-Spoofing Systems
Voice authentication has undergone significant changes from traditional systems that relied on handcrafted acoustic features to deep learning models that can extract robust speaker embeddings. This advancement has expanded its applications across finance, smart devices, law enforcement, and beyond. However, as adoption has grown, so have the threats. This survey presents a comprehensive review of the modern threat landscape targeting Voice Authentication Systems (VAS) and Anti-Spoofing Countermeasures (CMs), including data poisoning, adversarial, deepfake, and adversarial spoofing attacks. We chronologically trace the development of voice authentication and examine how vulnerabilities have evolved in tandem with technological advancements. For each category of attack, we summarize methodologies, highlight commonly used datasets, compare performance and limitations, and organize existing literature using widely accepted taxonomies. By highlighting emerging risks and open challenges, this survey aims to support the development of more secure and resilient voice authentication systems.
comment: This paper will be submitted to the Computer Science Review
HonestCyberEval: An AI Cyber Risk Benchmark for Automated Software Exploitation
We introduce HonestCyberEval, a new benchmark for assessing AI models' capabilities and risks in automated software exploitation, focusing on their ability to detect and exploit vulnerabilities in real-world software systems. Our evaluation leverages the Nginx web server repository augmented with synthetic vulnerabilities. We assess several leading language models, including OpenAI's GPT-4.5, o3-mini, o1 and o1-mini, Anthropic's Claude-3-7-sonnet-20250219, Claude-3.5-sonnet-20241022 and Claude-3.5-sonnet-20240620, Google DeepMind's Gemini-1.5-pro, and OpenAI's earlier GPT-4o model. Our findings reveal that these models vary significantly in their success rates and efficiency, with o1-preview achieving the highest success rate (92.85\%) and o3-mini and Claude-3.7-sonnet-20250219 providing cost-effective but less successful alternatives. This risk evaluation establishes a foundation for systematically evaluating the AI cyber risk in realistic cyber offence operations.
Fingerprint Vector: Enabling Scalable and Efficient Model Fingerprint Transfer via Vector Addition
Backdoor-based fingerprinting has emerged as an effective technique for tracing the ownership of large language models. However, in real-world deployment scenarios, developers often instantiate multiple downstream models from a shared base model, and applying fingerprinting to each variant individually incurs prohibitive computational overhead. While inheritance-based approaches -- where fingerprints are embedded into the base model and expected to persist through fine-tuning -- appear attractive, they suffer from three key limitations: late-stage fingerprinting, fingerprint instability, and interference with downstream adaptation. To address these challenges, we propose a novel mechanism called the Fingerprint Vector. Our method first embeds a fingerprint into the base model via backdoor-based fine-tuning, then extracts a task-specific parameter delta as a fingerprint vector by computing the difference between the fingerprinted and clean models. This vector can be directly added to any structurally compatible downstream model, allowing the fingerprint to be transferred post hoc without additional fine-tuning. Extensive experiments show that Fingerprint Vector achieves comparable or superior performance to direct injection across key desiderata. It maintains strong effectiveness across diverse model architectures as well as mainstream downstream variants within the same family. It also preserves harmlessness and robustness in most cases. Even when slight robustness degradation is observed, the impact remains within acceptable bounds and is outweighed by the scalability benefits of our approach.
Secure Reinforcement Learning via Shuffle Privacy Model
Reinforcement learning (RL) is a powerful tool for sequential decision-making, but its application is often hindered by privacy concerns arising from its interaction data. This challenge is particularly acute in advanced Cyber-Physical Systems (CPS), where learning from operational and user data can expose systems to privacy inference attacks. Existing differential privacy (DP) models for RL are often inadequate: the centralized model requires a fully trusted server, creating a single point of failure risk, while the local model incurs significant performance degradation that is unsuitable for many control applications. This paper addresses this gap by leveraging the emerging shuffle model of privacy, an intermediate trust model that provides strong privacy guarantees without a centralized trust assumption. We present Shuffle Differentially Private Policy Elimination (SDP-PE), the first generic policy elimination-based algorithm for episodic RL under the shuffle model. Our method introduces a novel exponential batching schedule and a ``forgetting'' mechanism to balance the competing demands of privacy and learning performance. Our analysis shows that SDP-PE achieves a near-optimal regret bound, demonstrating a superior privacy-regret trade-off that significantly outperforms the local model. This work establishes the viability of the shuffle model for secure data-driven control in advanced CPS.
sudoLLM: On Multi-role Alignment of Language Models EMNLP 2025
User authorization-based access privileges are a key feature in many safety-critical systems, but have not been extensively studied in the large language model (LLM) realm. In this work, drawing inspiration from such access control systems, we introduce sudoLLM, a novel framework that results in multi-role aligned LLMs, i.e., LLMs that account for, and behave in accordance with, user access rights. sudoLLM injects subtle user-based biases into queries and trains an LLM to utilize this bias signal in order to produce sensitive information if and only if the user is authorized. We present empirical results demonstrating that this approach shows substantially improved alignment, generalization, resistance to prefix-based jailbreaking attacks, and ``fails-closed''. The persistent tension between the language modeling objective and safety alignment, which is often exploited to jailbreak LLMs, is somewhat resolved with the aid of the injected bias signal. Our framework is meant as an additional security layer, and complements existing guardrail mechanisms for enhanced end-to-end safety with LLMs.
comment: Accepted to EMNLP 2025 (findings)
SoK: A Literature and Engineering Review of Regular Expression Denial of Service (ReDoS) CCS '25
Regular Expression Denial of Service (ReDoS) is a vulnerability class that has become prominent in recent years. Attackers can weaponize such weaknesses as part of asymmetric cyberattacks that exploit the slow worst-case matching time of regular expression (regex) engines. In the past, problematic regexes have led to outages at Cloudflare and Stack Overflow, showing the severity of the problem. While ReDoS has drawn significant research attention, there has been no systematization of knowledge to delineate the state of the art and identify opportunities for further research. In this paper, we describe the existing knowledge on ReDoS. We first provide a systematic literature review, discussing approaches for detecting, preventing, and mitigating ReDoS vulnerabilities. Then, our engineering review surveys the latest regex engines to examine whether and how ReDoS defenses have been realized. Combining our findings, we observe that (1) in the literature, almost no studies evaluate whether and how ReDoS vulnerabilities can be weaponized against real systems, making it difficult to assess their real-world impact; and (2) from an engineering view, many mainstream regex engines have introduced partial or full ReDoS defenses, rendering many threat models obsolete. We conclude by highlighting avenues for future work. The open challenges in ReDoS research are to evaluate emerging defenses and support engineers in migrating to defended engines. We also highlight the parallel between performance bugs and asymmetric DoS, and we argue that future work should capitalize more on this similarity and adopt a more systematic view on ReDoS-like vulnerabilities.
comment: 17 pages, 5 figures. Published in ASIA CCS '25
Breaking Data Silos: Towards Open and Scalable Mobility Foundation Models via Generative Continual Learning
Foundation models have revolutionized fields such as natural language processing and computer vision by enabling general-purpose learning across diverse tasks and datasets. However, building analogous models for human mobility remains challenging due to the privacy-sensitive nature of mobility data and the resulting data silos across institutions. To bridge this gap, we propose MoveGCL, a scalable and privacy-preserving framework for training mobility foundation models via generative continual learning. Without sharing raw data, MoveGCL enables decentralized and progressive model evolution by replaying synthetic trajectories generated from a frozen teacher model, and reinforces knowledge retention through a tailored distillation strategy that mitigates catastrophic forgetting. To address the heterogeneity of mobility patterns, MoveGCL incorporates a Mixture-of-Experts Transformer with a mobility-aware expert routing mechanism, and employs a layer-wise progressive adaptation strategy to stabilize continual updates. Experiments on six real-world urban datasets demonstrate that MoveGCL achieves performance comparable to joint training and significantly outperforms federated learning baselines, while offering strong privacy protection. MoveGCL marks a crucial step toward unlocking foundation models for mobility, offering a practical blueprint for open, scalable, and privacy-preserving model development in the era of foundation models. To facilitate reproducibility and future research, we have released the code and models at https://github.com/tsinghua-fib-lab/MoveGCL.
comment: The 33rd ACM International Conference on Advances in Geographic Information Systems
Can LLMs Handle WebShell Detection? Overcoming Detection Challenges with Behavioral Function-Aware Framework
WebShell attacks, where malicious scripts are injected into web servers, pose a significant cybersecurity threat. Traditional ML and DL methods are often hampered by challenges such as the need for extensive training data, catastrophic forgetting, and poor generalization. Recently, Large Language Models have emerged as powerful alternatives for code-related tasks, but their potential in WebShell detection remains underexplored. In this paper, we make two contributions: (1) a comprehensive evaluation of seven LLMs, including GPT-4, LLaMA 3.1 70B, and Qwen 2.5 variants, benchmarked against traditional sequence- and graph-based methods using a dataset of 26.59K PHP scripts, and (2) the Behavioral Function-Aware Detection (BFAD) framework, designed to address the specific challenges of applying LLMs to this domain. Our framework integrates three components: a Critical Function Filter that isolates malicious PHP function calls, a Context-Aware Code Extraction strategy that captures the most behaviorally indicative code segments, and Weighted Behavioral Function Profiling that enhances in-context learning by prioritizing the most relevant demonstrations based on discriminative function-level profiles. Our results show that, stemming from their distinct analytical strategies, larger LLMs achieve near-perfect precision but lower recall, while smaller models exhibit the opposite trade-off. However, all baseline models lag behind previous SOTA methods. With the application of BFAD, the performance of all LLMs improves significantly, yielding an average F1 score increase of 13.82%. Notably, larger models now outperform SOTA benchmarks, while smaller models such as Qwen-2.5-Coder-3B achieve performance competitive with traditional methods. This work is the first to explore the feasibility and limitations of LLMs for WebShell detection and provides solutions to address the challenges in this task.
comment: Published as a conference paper at COLM 2025
Unveiling the Landscape of LLM Deployment in the Wild: An Empirical Study
Large language models (LLMs) are increasingly deployed through open-source and commercial frameworks, enabling individuals and organizations to self-host advanced LLM capabilities. As LLM deployments become prevalent, particularly in industry, ensuring their secure and reliable operation has become a critical issue. However, insecure defaults and misconfigurations often expose LLM services to the public internet, posing serious security and system engineering risks. This study conducted a large-scale empirical investigation of public-facing LLM deployments, focusing on the prevalence of services, exposure characteristics, systemic vulnerabilities, and associated risks. Through internet-wide measurements, we identified 320,102 public-facing LLM services across 15 frameworks and extracted 158 unique API endpoints, categorized into 12 functional groups based on functionality and security risk. Our analysis found that over 40% of endpoints used plain HTTP, and over 210,000 endpoints lacked valid TLS metadata. API exposure was highly inconsistent: some frameworks, such as Ollama, responded to over 35% of unauthenticated API requests, with about 15% leaking model or system information, while other frameworks implemented stricter controls. We observed widespread use of insecure protocols, poor TLS configurations, and unauthenticated access to critical operations. These security risks, such as model leakage, system compromise, and unauthorized access, are pervasive and highlight the need for a secure-by-default framework and stronger deployment practices.
Optimal Planning for Enhancing the Resilience of Modern Distribution Systems Against Cyberattacks
The increasing integration of IoT-connected devices in smart grids has introduced new vulnerabilities at the distribution level. Of particular concern is the potential for cyberattacks that exploit high-wattage IoT devices, such as EV chargers, to manipulate local demand and destabilize the grid. While previous studies have primarily focused on such attacks at the transmission level, this paper investigates their feasibility and impact at the distribution level. We examine how cyberattackers can target voltage-sensitive nodes, especially those exposed by the presence of high-consumption devices, to cause voltage deviation and service disruption. Our analysis demonstrates that conventional grid protections are insufficient against these intelligent, localized attacks. To address this, we propose resilience strategies using distributed generation (DGs), exploring their role in preemptive planning. This research highlights the urgent need for distribution-level cyber resilience planning in smart grids.
comment: Accepted Paper
Large Language Model-Based Framework for Explainable Cyberattack Detection in Automatic Generation Control Systems
The increasing digitization of smart grids has improved operational efficiency but also introduced new cybersecurity vulnerabilities, such as False Data Injection Attacks (FDIAs) targeting Automatic Generation Control (AGC) systems. While machine learning (ML) and deep learning (DL) models have shown promise in detecting such attacks, their opaque decision-making limits operator trust and real-world applicability. This paper proposes a hybrid framework that integrates lightweight ML-based attack detection with natural language explanations generated by Large Language Models (LLMs). Classifiers such as LightGBM achieve up to 95.13% attack detection accuracy with only 0.004 s inference latency. Upon detecting a cyberattack, the system invokes LLMs, including GPT-3.5 Turbo, GPT-4 Turbo, and GPT-4o mini, to generate human-readable explanation of the event. Evaluated on 100 test samples, GPT-4o mini with 20-shot prompting achieved 93% accuracy in identifying the attack target, a mean absolute error of 0.075 pu in estimating attack magnitude, and 2.19 seconds mean absolute error (MAE) in estimating attack onset. These results demonstrate that the proposed framework effectively balances real-time detection with interpretable, high-fidelity explanations, addressing a critical need for actionable AI in smart grid cybersecurity.
comment: Accepted Paper
Cryptography and Security 13
Collaborative Intelligence: Topic Modelling of Large Language Model use in Live Cybersecurity Operations
Objective: This work describes the topic modelling of Security Operations Centre (SOC) use of a large language model (LLM), during live security operations. The goal is to better understand how these specialists voluntarily use this tool. Background: Human-automation teams have been extensively studied, but transformer-based language models have sparked a new wave of collaboration. SOC personnel at a major cybersecurity provider used an LLM to support live security operations. This study examines how these specialists incorporated the LLM into their work. Method: Our data set is the result of 10 months of SOC operators accessing GPT-4 over an internally deployed HTTP-based chat application. We performed two topic modelling exercises, first using the established BERTopic model (Grootendorst, 2022), and second, using a novel topic modeling workflow. Results: Both the BERTopic analysis and novel modelling approach revealed that SOC operators primarily used the LLM to facilitate their understanding of complex text strings. Variations on this use-case accounted for ~40% of SOC LLM usage. Conclusion: SOC operators are required to rapidly interpret complex commands and similar information. Their natural tendency to leverage LLMs to support this activity indicates that their workflow can be supported and augmented by designing collaborative LLM tools for use in the SOC. Application: This work can aid in creating next-generation tools for Security Operations Centres. By understanding common use-cases, we can develop workflows supporting SOC task flow. One example is a right-click context menu for executing a command line analysis LLM call directly in the SOC environment.
An 8- and 12-bit block AES cipher
Because it is so unusual, or hard to find, or expository, a truly tiny 8- or 12-bit block AES (Rijndael) cipher is documented here, along with Java source code.
comment: This "research note" of mine from 2013 has been requested so often from me over the years, along with requests for a way to cite it properly, that I think it's appropriate to put it out on the web in a citeable archive. Arxiv, step up
Privacy-Preserving Federated Learning Framework for Risk-Based Adaptive Authentication
Balancing robust security with strong privacy guarantees is critical for Risk-Based Adaptive Authentication (RBA), particularly in decentralized settings. Federated Learning (FL) offers a promising solution by enabling collaborative risk assessment without centralizing user data. However, existing FL approaches struggle with Non-Independent and Identically Distributed (Non-IID) user features, resulting in biased, unstable, and poorly generalized global models. This paper introduces FL-RBA2, a novel Federated Learning framework for Risk-Based Adaptive Authentication that addresses Non-IID challenges through a mathematically grounded similarity transformation. By converting heterogeneous user features (including behavioral, biometric, contextual, interaction-based, and knowledge-based modalities) into IID similarity vectors, FL-RBA2 supports unbiased aggregation and personalized risk modeling across distributed clients. The framework mitigates cold-start limitations via clustering-based risk labeling, incorporates Differential Privacy (DP) to safeguard sensitive information, and employs Message Authentication Codes (MACs) to ensure model integrity and authenticity. Federated updates are securely aggregated into a global model, achieving strong balance between user privacy, scalability, and adaptive authentication robustness. Rigorous game-based security proofs in the Random Oracle Model formally establish privacy, correctness, and adaptive security guarantees. Extensive experiments on keystroke, mouse, and contextual datasets validate FL-RBA2's effectiveness in high-risk user detection and its resilience to model inversion and inference attacks, even under strong DP constraints.
A Systematic Approach to Predict the Impact of Cybersecurity Vulnerabilities Using LLMs
Vulnerability databases, such as the National Vulnerability Database (NVD), offer detailed descriptions of Common Vulnerabilities and Exposures (CVEs), but often lack information on their real-world impact, such as the tactics, techniques, and procedures (TTPs) that adversaries may use to exploit the vulnerability. However, manually linking CVEs to their corresponding TTPs is a challenging and time-consuming task, and the high volume of new vulnerabilities published annually makes automated support desirable. This paper introduces TRIAGE, a two-pronged automated approach that uses Large Language Models (LLMs) to map CVEs to relevant techniques from the ATT&CK knowledge base. We first prompt an LLM with instructions based on MITRE's CVE Mapping Methodology to predict an initial list of techniques. This list is then combined with the results from a second LLM-based module that uses in-context learning to map a CVE to relevant techniques. This hybrid approach strategically combines rule-based reasoning with data-driven inference. Our evaluation reveals that in-context learning outperforms the individual mapping methods, and the hybrid approach improves recall of exploitation techniques. We also find that GPT-4o-mini performs better than Llama3.3-70B on this task. Overall, our results show that LLMs can be used to automatically predict the impact of cybersecurity vulnerabilities and TRIAGE makes the process of mapping CVEs to ATT&CK more efficient. Keywords: vulnerability impact, CVE, ATT&CK techniques, large language models, automated mapping.
Securing Face and Fingerprint Templates in Humanitarian Biometric Systems
In humanitarian and emergency scenarios, the use of biometrics can dramatically improve the efficiency of operations, but it poses risks for the data subjects, which are exacerbated in contexts of vulnerability. To address this, we present a mobile biometric system implementing a biometric template protection (BTP) scheme suitable for these scenarios. After rigorously formulating the functional, operational, and security and privacy requirements of these contexts, we perform a broad comparative analysis of the BTP landscape. PolyProtect, a method designed to operate on neural network face embeddings, is identified as the most suitable method due to its effectiveness, modularity, and lightweight computational burden. We evaluate PolyProtect in terms of verification and identification accuracy, irreversibility, and unlinkability, when this BTP method is applied to face embeddings extracted using EdgeFace, a novel state-of-the-art efficient feature extractor, on a real-world face dataset from a humanitarian field project in Ethiopia. Moreover, as PolyProtect promises to be modality-independent, we extend its evaluation to fingerprints. To the best of our knowledge, this is the first time that PolyProtect has been evaluated for the identification scenario and for fingerprint biometrics. Our experimental results are promising, and we plan to release our code
Training Language Model Agents to Find Vulnerabilities with CTF-Dojo
Large language models (LLMs) have demonstrated exceptional capabilities when trained within executable runtime environments, notably excelling at software engineering tasks through verified feedback loops. Yet, scalable and generalizable execution-grounded environments remain scarce, limiting progress in training more capable ML agents. We introduce CTF-Dojo, the first large-scale executable runtime tailored for training LLMs with verifiable feedback, featuring 658 fully functional Capture-The-Flag (CTF)-style challenges containerized in Docker with guaranteed reproducibility. To enable rapid scaling without manual intervention, we develop CTF-Forge, an automated pipeline that transforms publicly available artifacts into ready-to-use execution environments in minutes, eliminating weeks of expert configuration traditionally required. We trained LLM-based agents on just 486 high-quality, execution-verified trajectories from CTF-Dojo, achieving up to 11.6% absolute gains over strong baselines across three competitive benchmarks: InterCode-CTF, NYU CTF Bench, and Cybench. Our best-performing 32B model reaches 31.9% Pass@1, establishing a new open-weight state-of-the-art that rivals frontier models like DeepSeek-V3-0324 and Gemini-2.5-Flash. By framing CTF-style tasks as a benchmark for executable-agent learning, CTF-Dojo demonstrates that execution-grounded training signals are not only effective but pivotal in advancing high-performance ML agents without dependence on costly proprietary systems.
Learning from Few Samples: A Novel Approach for High-Quality Malcode Generation
Intrusion Detection Systems (IDS) play a crucial role in network security defense. However, a significant challenge for IDS in training detection models is the shortage of adequately labeled malicious samples. To address these issues, this paper introduces a novel semi-supervised framework \textbf{GANGRL-LLM}, which integrates Generative Adversarial Networks (GANs) with Large Language Models (LLMs) to enhance malicious code generation and SQL Injection (SQLi) detection capabilities in few-sample learning scenarios. Specifically, our framework adopts a collaborative training paradigm where: (1) the GAN-based discriminator improves malicious pattern recognition through adversarial learning with generated samples and limited real samples; and (2) the LLM-based generator refines the quality of malicious code synthesis using reward signals from the discriminator. The experimental results demonstrate that even with a limited number of labeled samples, our training framework is highly effective in enhancing both malicious code generation and detection capabilities. This dual enhancement capability offers a promising solution for developing adaptive defense systems capable of countering evolving cyber threats.
comment: 18pages,5 figures,emnlp
Client-Aided Secure Two-Party Computation of Dynamic Controllers
In this paper, we propose a secure two-party computation protocol for dynamic controllers using a secret sharing scheme. The proposed protocol realizes outsourcing of controller computation to two servers, while controller parameters, states, inputs, and outputs are kept secret against the servers. Unlike previous encrypted controls in a single-server setting, the proposed method can operate a dynamic controller for an infinite time horizon without controller state decryption or input re-encryption. We show that the control performance achievable by the proposed protocol can be made arbitrarily close to that attained by the unencrypted controller. Furthermore, system-theoretic and cryptographic modifications of the protocol are presented to improve the communication complexity. The feasibility of the protocol is demonstrated through numerical examples of PID and observer-based controls.
comment: 12 pages, 4 figures
Prefill-level Jailbreak: A Black-Box Risk Analysis of Large Language Models
Large Language Models face security threats from jailbreak attacks. Existing research has predominantly focused on prompt-level attacks while largely ignoring the underexplored attack surface of user-controlled response prefilling. This functionality allows an attacker to dictate the beginning of a model's output, thereby shifting the attack paradigm from persuasion to direct state manipulation.In this paper, we present a systematic black-box security analysis of prefill-level jailbreak attacks. We categorize these new attacks and evaluate their effectiveness across fourteen language models. Our experiments show that prefill-level attacks achieve high success rates, with adaptive methods exceeding 99% on several models. Token-level probability analysis reveals that these attacks work through initial-state manipulation by changing the first-token probability from refusal to compliance.Furthermore, we show that prefill-level jailbreak can act as effective enhancers, increasing the success of existing prompt-level attacks by 10 to 15 percentage points. Our evaluation of several defense strategies indicates that conventional content filters offer limited protection. We find that a detection method focusing on the manipulative relationship between the prompt and the prefill is more effective. Our findings reveal a gap in current LLM safety alignment and highlight the need to address the prefill attack surface in future safety training.
Blockchain Security Risk Assessment in Quantum Era, Migration Strategies and Proactive Defense
The emergence of quantum computing presents a formidable challenge to the security of blockchain systems. Traditional cryptographic algorithms, foundational to digital signatures, message encryption, and hashing functions, become vulnerable to the immense computational power of quantum computers. This paper conducts a thorough risk assessment of transitioning to quantum-resistant blockchains, comprehensively analyzing potential threats targeting vital blockchain components: the network, mining pools, transaction verification mechanisms, smart contracts, and user wallets. By elucidating the intricate challenges and strategic considerations inherent in transitioning to quantum-resistant algorithms, the paper evaluates risks and highlights obstacles in securing blockchain components with quantum-resistant cryptography. It offers a hybrid migration strategy to facilitate a smooth transition from classical to quantum-resistant cryptography. The analysis extends to prominent blockchains such as Bitcoin, Ethereum, Ripple, Litecoin, and Zcash, assessing vulnerable components, potential impacts, and associated STRIDE threats, thereby identifying areas susceptible to quantum attacks. Beyond analysis, the paper provides actionable guidance for designing secure and resilient blockchain ecosystems in the quantum computing era. Recognizing the looming threat of quantum computers, this research advocates for a proactive transition to quantum-resistant blockchain networks. It proposes a tailored security blueprint that strategically fortifies each component against the evolving landscape of quantum-induced cyber threats. Emphasizing the critical need for blockchain stakeholders to adopt proactive measures and implement quantum-resistant solutions, the paper underscores the importance of embracing these insights to navigate the complexities of the quantum era with resilience and confidence.
Cyber-Zero: Training Cybersecurity Agents without Runtime
Large Language Models (LLMs) have achieved remarkable success in software engineering tasks when trained with executable runtime environments, particularly in resolving GitHub issues. However, such runtime environments are often unavailable in other domains, especially cybersecurity, where challenge configurations and execution contexts are ephemeral or restricted. We present Cyber-Zero, the first runtime-free framework for synthesizing high-quality agent trajectories to train cybersecurity LLMs. Cyber-Zero leverages publicly available CTF writeups and employs persona-driven LLM simulation to reverse-engineer runtime behaviors and generate realistic, long-horizon interaction sequences without actual environments. Using trajectories synthesized by Cyber-Zero, we train LLM-based agents that achieve up to 13.1% absolute performance gains over baseline models on three prominent CTF benchmarks: InterCode-CTF, NYU CTF Bench, and Cybench. Our best model, Cyber-Zero-32B, establishes new state-of-the-art performance among open-weight models, matching the capabilities of proprietary systems like DeepSeek-V3-0324 and Claude-3.5-Sonnet while offering superior cost-effectiveness, and demonstrating that runtime-free trajectory synthesis can effectively democratize the development of state-of-the-art cybersecurity agents.
comment: Public Link: https://github.com/amazon-science/cyber-zero
Defending Against Prompt Injection With a Few DefensiveTokens
When large language model (LLM) systems interact with external data to perform complex tasks, a new attack, namely prompt injection, becomes a significant threat. By injecting instructions into the data accessed by the system, the attacker is able to override the initial user task with an arbitrary task directed by the attacker. To secure the system, test-time defenses, e.g., defensive prompting, have been proposed for system developers to attain security only when needed in a flexible manner. However, they are much less effective than training-time defenses that change the model parameters. Motivated by this, we propose DefensiveToken, a test-time defense with prompt injection robustness comparable to training-time alternatives. DefensiveTokens are newly inserted as special tokens, whose embeddings are optimized for security. In security-sensitive cases, system developers can append a few DefensiveTokens before the LLM input to achieve security with a minimal utility drop. In scenarios where security is less of a concern, developers can simply skip DefensiveTokens; the LLM system remains the same as there is no defense, generating high-quality responses. Thus, DefensiveTokens, if released alongside the model, allow a flexible switch between the state-of-the-art (SOTA) utility and almost-SOTA security at test time. The code is available at https://github.com/Sizhe-Chen/DefensiveToken.
Confidential Prompting: Privacy-preserving LLM Inference on Cloud
This paper introduces a vision of confidential prompting: securing user prompts from untrusted, cloud-hosted large language model (LLM) provider while preserving model confidentiality, output invariance, and compute efficiency. As a first step toward this vision, we present Obfuscated Secure Partitioned Decoding (OSPD), a system built on two key innovations. First, Secure Partitioned Decoding (SPD) isolates user prompts within per-user processes residing in a confidential virtual machine (CVM) on the cloud, which are inaccessible for the cloud LLM while allowing it to generate tokens efficiently. Second, Prompt Obfuscation (PO) introduces a novel cryptographic technique that enhances SPD resilience against advanced prompt reconstruction attacks. Together, these innovations ensure OSPD protects both prompt and model confidentiality while maintaining service functionality. OSPD enables practical, privacy-preserving cloud-hosted LLM inference for sensitive applications, such as processing personal data, clinical records, and financial documents.