arXiv subjects

Cong Wu

Publications and source records attributed to Cong Wu.

At least 19 recordsLinked to original sources

arXiv2026

MoE Proxy Models for Low-Cost Failure Reproduction and Diagnosis in LLM RL Post-Training

Reinforcement learning (RL) post-training of large language models (LLMs) is computationally intensive and involves complex system pipelines with substantial debugging overhead. In practice, factors such as framework adaptation, numerical precision, and operator implementation can cause failures, including gradient overflow and loss divergence. Reproducing such failures directly on large models requires considerable time and computational resources. This paper systematically analyzes failures encountered during large-scale RL training on the Huawei Ascend platform, summarizes representative failure types, and identifies three model-side factors relevant to fault reproduction. Based on these factors, we propose a proxy-model construction method for low-cost fault investigation and auxiliary diagnosis. It employs structure-preserving, clustering-based expert pruning to select representative experts while retaining the model's backbone architecture, routing mechanism, and basic task capabilities. Our experimental results show that the proxy models reduce accelerator requirements by 50%-87.5% and achieve up to a 33.3x reduction in per-step NPU-hour cost, while preserving major training dynamics and reproducing fault responses consistent with the original models. Overall, the proxy models can serve as low-cost surrogates for fault reproduction, targeted validation, and auxiliary diagnosis in RL post-training.

cs.LG↗

arXiv2026

Babel: Jailbreaking Safety Attention via Obfuscation Distribution Optimized Sampling

Despite rigorous safety alignment, Large Language Models (LLMs) remain vulnerable to jailbreak attacks. Existing black-box methods often rely on heuristic templates or exhaustive trials, lacking mechanistic interpretability and query efficiency. In this study, we investigate an intrinsic vulnerability in the safety mechanisms of LLMs, where safety alignment relies on a small set of sparsely distributed attention heads, leaving much of the representational space weakly monitored. We formalize this phenomenon with a mathematical jailbreaking model that characterizes the delicate boundary of effective text obfuscation and analytically explains observed jailbreak behaviors. Guided by this model, we propose Babel, an efficient black-box attack framework that exploits the identified safety gap through systematic obfuscation sampling with iterative, feedback-driven distribution refinement, enabling reliable and high-success jailbreak attacks without access to model internals. Comprehensive evaluations on frontier commercial models demonstrate that Babel achieves state-of-the-art attack success rates and superior query efficiency. Specifically, compared to state-of-the-art methods, Babel increases the attack success rate on GPT-4o from 41.33% to 82.67% and on Claude-3-5-haiku from 38.33% to 78.33% within an average of 40 queries, providing a robust red-teaming methodology for LLMs safety research.

cs.CR↗

arXiv2026

EvoPoC: Automated Exploit Synthesis for DeFi Smart Contracts via Hierarchical Knowledge Graphs

Smart contract vulnerabilities in Decentralized Finance caused over billions of dollars losses every year, yet the security community faces a critical bottleneck: identifying a vulnerability is not the same as proving it is exploitable. Manual PoC construction is prohibitively labor-intensive, leaving most disclosed vulnerabilities unverified and protocols exposed long before mitigation is applied. In this paper, we propose \sys, a knowledge-driven agentic system for end-to-end contract vulnerability detection and exploit synthesis. Our core insight is that exploit synthesis is not a code generation task but a \emph{structured reasoning problem} that requires grounded knowledge of protocol semantics, failure root cause, and exploit primitives. \sys organizes this knowledge into a \emph{Hierarchical Knowledge Graph} (HKG) that serves as structured memory for LLM-guided multi-hop reasoning. To validate exploit feasibility beyond code synthesis, \sys employs a two-stage validation framework that checks exploit-path reachability via SMT solving and profit realizability via asset-level state simulation, ensuring generated PoCs satisfy both logical and economic viability constraints. Evaluated on 88 real-world DeFi attacks and 72 audited projects (2,573 contracts), \sys achieves 98\% recall and 0.9 F1-score in detection, and a 96.6\% exploit success rate (ESR), reproducing 85 historical exploits and recovering over \$116.2M revenue. \sys outperforms SOTA fuzzers (\textsc{Verite}, \textsc{ItyFuzz}) by up to $5\times$ in ESR and $300\times$ in recoverable value, and the LLM-based exploit generator \textsc{A1} by $2\times$ and $8.5\times$ respectively. In bug bounty evaluation, \sys identified 16 confirmed 0-day vulnerabilities, helping secure over \$70.6M and earning \$2,900 in bounties.

cs.CR↗

arXiv2026

Phishing Detection in Ethereum via Temporal Graph Contrastive Learning

Blockchain and decentralized finance have revolutionized the financial ecosystem while simultaneously exposing it to cryptocurrency phishing attacks. Existing phishing detection methods primarily rely on graph learning, but they face significant limitations. Static graph learning approaches fail to account for the temporal evolution of phishing patterns, while semi-dynamic methods, such as those combining static GNNs with LSTM, struggle to capture the irregular and bursty nature of blockchain transactions. Moreover, these methods overlook the diversity of Ethereum transactions, treating them as homogeneous graphs, and heavily rely on supervised learning, which requires extensive labeled data that is not readily available. These limitations reduce their adaptability to emerging phishing threats. In this paper, we present PhishEye, a fully dynamic self-supervised system that monitors on-chain transactions to detect phishing activities. PhishEye formulates Ethereum transactions as a heterogeneous temporal attributed multi-graph and incorporates a novel temporal graph contrastive learning model, which captures both temporal patterns and heterogeneous transaction types. The evaluation on a dataset of 161,658 addresses and 416,541 transactions shows that PhishEye outperforms existing methods, achieving an F1 score of 87.23% and an AUC of 98.43% for phishing transaction detection, and an F1 score of 94.19% and an AUC of 98.03% for phishing account detection. In real-world deployment from May 1, 2023 to July 31, 2024, PhishEye identified 1,803 previously unknown phishing addresses, providing early alerts that helped prevent losses exceeding 2 billion USD.

cs.CR↗

arXiv2026

NTIRE 2026 Challenge on Video Saliency Prediction: Methods and Results

This paper presents an overview of the NTIRE 2026 Challenge on Video Saliency Prediction. The goal of the challenge participants was to develop automatic saliency map prediction methods for the provided video sequences. The novel dataset of 2,000 diverse videos with an open license was prepared for this challenge. The fixations and corresponding saliency maps were collected using crowdsourced mouse tracking and contain viewing data from over 5,000 assessors. Evaluation was performed on a subset of 800 test videos using generally accepted quality metrics. The challenge attracted over 20 teams making submissions, and 7 teams passed the final phase with code review. All data used in this challenge is made publicly available - https://github.com/msu-video-group/NTIRE26_Saliency_Prediction.

cs.CV↗

arXiv2026

CAAP: Capture-Aware Adversarial Patch Attacks on Palmprint Recognition Models

Palmprint recognition is deployed in security-critical applications, including access control and palm-based payment, due to its contactless acquisition and highly discriminative ridge-and-crease textures. However, the robustness of deep palmprint recognition systems against physically realizable attacks remains insufficiently understood. Existing studies are largely confined to the digital setting and do not adequately account for the texture-dominant nature of palmprint recognition or the distortions introduced during physical acquisition. To address this gap, we propose CAAP, a capture-aware adversarial patch framework for palmprint recognition. CAAP learns a universal patch that can be reused across inputs while remaining effective under realistic acquisition variation. To match the structural characteristics of palmprints, the framework adopts a cross-shaped patch topology, which enlarges spatial coverage under a fixed pixel budget and more effectively disrupts long-range texture continuity. CAAP further integrates three modules: ASIT for input-conditioned patch rendering, RaS for stochastic capture-aware simulation, and MS-DIFE for feature-level identity-disruptive guidance. We evaluate CAAP on the Tongji, IITD, and AISEC datasets against generic CNN backbones and palmprint-specific recognition models. Experiments show that CAAP achieves strong untargeted and targeted attack performance with favorable cross-model and cross-dataset transferability. The results further show that, although adversarial training can partially reduce the attack success rate, substantial residual vulnerability remains. These findings indicate that deep palmprint recognition systems remain vulnerable to physically realizable, capture-aware adversarial patch attacks, underscoring the need for more effective defenses in practice. Code available at https://github.com/ryliu68/CAAP.

cs.CV↗

arXiv2026

Decomposing Centrally Symmetric Convex Polyhedral Surfaces into Parallelograms

Let $\mathcal{M}_{2N}(\delta_1, \delta_2,\dots, \delta_N)$ be the moduli space of centrally symmetric convex polyhedral surfaces with $2N$ labeled vertices and prescribed cone-deficits $\delta_1$, $\delta_2$, $\dots$, $\delta_N$. We show that $\mathcal{M}_{2N}(\delta_1, \delta_2,\dots, \delta_N)$ has the structure of a real hyperbolic manifold of dimension $2N-3$. When $N=4$ and $5$, we show that every surface in $\mathcal{M}_{2N}(\delta_1, \delta_2,\dots, \delta_N)$ can be decomposed into at most $2\binom{2N-2}{2}$ parallelograms, and the decomposition is invariant under the antipodal map. Using the edge-lengths of these parallelograms as coordinates, we show that the moduli space of centrally symmetric polyhedral surfaces with $8$ unlabeled vertices and cone-deficits $\frac{\pi}{2}$ is isometric to the quotient of a real hyperbolic regular ideal $5$-simplex by the dihedral group $D_6$.

math.GT↗

arXiv2026

SplitCom: Communication-efficient Split Federated Fine-tuning of LLMs via Temporal Compression

Federated fine-tuning of on-device large language models (LLMs) mitigates privacy concerns by preventing raw data sharing. However, the intensive computational and memory demands pose significant challenges for resource-constrained edge devices. To overcome these limitations, split federated learning (SFL) emerges as a promising solution that partitions the model into lightweight client-side and compute-intensive server-side sub-models, thus offloading the primary training workload to a powerful server. Nevertheless, high-dimensional activation exchanges in SFL lead to excessive communication overhead. To overcome this, we propose SplitCom, a communication-efficient SFL framework for LLMs that exploits temporal redundancy in activations across consecutive training epochs. Inspired by video compression, the core innovation of our framework lies in selective activation uploading only when a noticeable deviation from previous epochs occurs. To balance communication efficiency and learning performance, we introduce two adaptive threshold control schemes based on 1) bang-bang control or 2) deep deterministic policy gradient (DDPG)-based reinforcement learning. Moreover, we implement dimensionality reduction techniques to alleviate client-side memory requirements. Furthermore, we extend SplitCom to the U-shape architecture, ensuring the server never accesses clients' labels. Extensive simulations and laboratory experiments demonstrate that SplitCom reduces uplink communication costs by up to 98.6\,\% in its standard configuration and total communication costs by up to 95.8\,\% in its U-shape variant without noticeably compromising model performance.

cs.NI↗

arXiv2026

RECUR: Resource Exhaustion Attack via Recursive-Entropy Guided Counterfactual Utilization and Reflection

Large Reasoning Models (LRMs) employ reasoning to address complex tasks. Such explicit reasoning requires extended context lengths, resulting in substantially higher resource consumption. Prior work has shown that adversarially crafted inputs can trigger redundant reasoning processes, exposing LRMs to resource-exhaustion vulnerabilities. However, the reasoning process itself, especially its reflective component, has received limited attention, even though it can lead to over-reflection and consume excessive computing power. In this paper, we introduce Recursive Entropy to quantify the risk of resource consumption in reflection, thereby revealing the safety issues inherent in inference itself. Based on Recursive Entropy, we introduce RECUR, a resource exhaustion attack via Recursive Entropy guided Counterfactual Utilization and Reflection. It constructs counterfactual questions to verify the inherent flaws and risks of LRMs. Extensive experiments demonstrate that, under benign inference, recursive entropy exhibits a pronounced decreasing trend. RECUR disrupts this trend, increasing the output length by up to 11x and decreasing throughput by 90%. Our work provides a new perspective on robust reasoning.

cs.AI↗

arXiv2025

Don't Trust Your Upstream: Exploiting LLM Multi-Agent System via Topology-Guided Adversarial Propagation

The digital world is witnessing the rapid rise of LLM-based multi-agent systems (MASs) and their powerful applications. However, their security remains insufficiently understood, as existing evaluations are largely limited to narrow attack settings and may substantially underestimate the real risks of MAS deployments. Inspired by the MAS inter-agent dependencies, where upstream outputs are reinterpreted and executed by downstream agents, we propose a topology-aware attack scheme that propagates adversarial contamination from exposed edge agents to high-privilege agents to induce malicious behaviors. By combining topology reconnaissance, contamination propagation modeling, and hierarchical payload encapsulation, our approach overcomes the key challenges of black-box attacks and makes such multi-hop compromise practical. Experiments show that our approach achieves success rates of 40\%--78\% on three widely-used MAS frameworks under five topologies, and 85\% on two real-world MAS applications across 20 representative scenarios. The results reveal fundamental vulnerabilities in MASs that have been overlooked by prior studies. Based on these findings, we propose a topology-trust mitigation that blocks 94.8\% of such composite attacks.

cs.CR↗

arXiv2025

Investigating Sharding Advancements, Methodologies, and Adoption Potential in Hedera

Sharding has emerged as a critical solution to address the scalability challenges faced by blockchain networks, enabling them to achieve higher transaction throughput, reduced latency, and optimized resource usage. This paper investigates the advancements, methodologies, and adoption potential of sharding in the context of Hedera, a distributed ledger technology known for its unique Gossip about Gossip protocol and asynchronous Byzantine Fault Tolerance (ABFT). We explore various academic and industrial sharding techniques, emphasizing their benefits and trade-offs. Building on these insights, we propose a hybrid sharding solution for Hedera that partitions the network into local and global committees, facilitating efficient cross-shard transactions and ensuring robust security through dynamic reconfiguration. Our analysis highlights significant reductions in storage and communication overhead, improved scalability, and enhanced fault tolerance, demonstrating the feasibility and advantages of integrating sharding into Hedera's architecture.

cs.DC↗

arXiv2025

Defensive Adversarial CAPTCHA: A Semantics-Driven Framework for Natural Adversarial Example Generation

Traditional CAPTCHA (Completely Automated Public Turing Test to Tell Computers and Humans Apart) schemes are increasingly vulnerable to automated attacks powered by deep neural networks (DNNs). Existing adversarial attack methods often rely on the original image characteristics, resulting in distortions that hinder human interpretation and limit their applicability in scenarios where no initial input images are available. To address these challenges, we propose the Unsourced Adversarial CAPTCHA (DAC), a novel framework that generates high-fidelity adversarial examples guided by attacker-specified semantics information. Leveraging a Large Language Model (LLM), DAC enhances CAPTCHA diversity and enriches the semantic information. To address various application scenarios, we examine the white-box targeted attack scenario and the black box untargeted attack scenario. For target attacks, we introduce two latent noise variables that are alternately guided in the diffusion step to achieve robust inversion. The synergy between gradient guidance and latent variable optimization achieved in this way ensures that the generated adversarial examples not only accurately align with the target conditions but also achieve optimal performance in terms of distributional consistency and attack effectiveness. In untargeted attacks, especially for black-box scenarios, we introduce bi-path unsourced adversarial CAPTCHA (BP-DAC), a two-step optimization strategy employing multimodal gradients and bi-path optimization for efficient misclassification. Experiments show that the defensive adversarial CAPTCHA generated by BP-DAC is able to defend against most of the unknown models, and the generated CAPTCHA is indistinguishable to both humans and DNNs.

cs.CV↗

arXiv2025

DP-TRAE: A Dual-Phase Merging Transferable Reversible Adversarial Example for Image Privacy Protection

In the field of digital security, Reversible Adversarial Examples (RAE) combine adversarial attacks with reversible data hiding techniques to effectively protect sensitive data and prevent unauthorized analysis by malicious Deep Neural Networks (DNNs). However, existing RAE techniques primarily focus on white-box attacks, lacking a comprehensive evaluation of their effectiveness in black-box scenarios. This limitation impedes their broader deployment in complex, dynamic environments. Further more, traditional black-box attacks are often characterized by poor transferability and high query costs, significantly limiting their practical applicability. To address these challenges, we propose the Dual-Phase Merging Transferable Reversible Attack method, which generates highly transferable initial adversarial perturbations in a white-box model and employs a memory augmented black-box strategy to effectively mislead target mod els. Experimental results demonstrate the superiority of our approach, achieving a 99.0% attack success rate and 100% recovery rate in black-box scenarios, highlighting its robustness in privacy protection. Moreover, we successfully implemented a black-box attack on a commercial model, further substantiating the potential of this approach for practical use.

cs.CR↗

arXiv2025

A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment

The remarkable success of Large Language Models (LLMs) has illuminated a promising pathway toward achieving Artificial General Intelligence for both academic and industrial communities, owing to their unprecedented performance across various applications. As LLMs continue to gain prominence in both research and commercial domains, their security and safety implications have become a growing concern, not only for researchers and corporations but also for every nation. Currently, existing surveys on LLM safety primarily focus on specific stages of the LLM lifecycle, e.g., deployment phase or fine-tuning phase, lacking a comprehensive understanding of the entire "lifechain" of LLMs. To address this gap, this paper introduces, for the first time, the concept of "full-stack" safety to systematically consider safety issues throughout the entire process of LLM training, deployment, and eventual commercialization. Compared to the off-the-shelf LLM safety surveys, our work demonstrates several distinctive advantages: (I) Comprehensive Perspective. We define the complete LLM lifecycle as encompassing data preparation, pre-training, post-training, deployment and final commercialization. To our knowledge, this represents the first safety survey to encompass the entire lifecycle of LLMs. (II) Extensive Literature Support. Our research is grounded in an exhaustive review of over 800+ papers, ensuring comprehensive coverage and systematic organization of security issues within a more holistic understanding. (III) Unique Insights. Through systematic literature analysis, we have developed reliable roadmaps and perspectives for each chapter. Our work identifies promising research directions, including safety in data generation, alignment techniques, model editing, and LLM-based agent systems. These insights provide valuable guidance for researchers pursuing future work in this field.

cs.CR↗

arXiv2025

CSI2Dig: Recovering Digit Content from Smartphone Loudspeakers Using Channel State Information

Eavesdropping on sounds emitted by mobile device loudspeakers can capture sensitive digital information, such as SMS verification codes, credit card numbers, and withdrawal passwords, which poses significant security risks. Existing schemes either require expensive specialized equipment, rely on spyware, or are limited to close-range signal acquisition. In this paper, we propose a scheme, CSI2Dig, for recovering digit content from Channel State Information (CSI) when digits are played through a smartphone loudspeaker. We observe that the electromagnetic interference caused by the audio signals from the loudspeaker affects the WiFi signals emitted by the phone's WiFi antenna. Building upon contrastive learning and denoising autoencoders, we develop a two-branch autoencoder network designed to amplify the impact of this electromagnetic interference on CSI. For feature extraction, we introduce the TS-Net, a model that captures relevant features from both the temporal and spatial dimensions of the CSI data. We evaluate our scheme across various devices, distances, volumes, and other settings. Experimental results demonstrate that our scheme can achieve an accuracy of 72.97%.

cs.CR↗

arXiv2025

PATFinger: Prompt-Adapted Transferable Fingerprinting against Unauthorized Multimodal Dataset Usage

The multimodal datasets can be leveraged to pre-train large-scale vision-language models by providing cross-modal semantics. Current endeavors for determining the usage of datasets mainly focus on single-modal dataset ownership verification through intrusive methods and non-intrusive techniques, while cross-modal approaches remain under-explored. Intrusive methods can adapt to multimodal datasets but degrade model accuracy, while non-intrusive methods rely on label-driven decision boundaries that fail to guarantee stable behaviors for verification. To address these issues, we propose a novel prompt-adapted transferable fingerprinting scheme from a training-free perspective, called PATFinger, which incorporates the global optimal perturbation (GOP) and the adaptive prompts to capture dataset-specific distribution characteristics. Our scheme utilizes inherent dataset attributes as fingerprints instead of compelling the model to learn triggers. The GOP is derived from the sample distribution to maximize embedding drifts between different modalities. Subsequently, our PATFinger re-aligns the adaptive prompt with GOP samples to capture the cross-modal interactions on the carefully crafted surrogate model. This allows the dataset owner to check the usage of datasets by observing specific prediction behaviors linked to the PATFinger during retrieval queries. Extensive experiments demonstrate the effectiveness of our scheme against unauthorized multimodal dataset usage on various cross-modal retrieval architectures by 30% over state-of-the-art baselines.

cs.IR↗

arXiv2025

\textsc{Perseus}: Tracing the Masterminds Behind Cryptocurrency Pump-and-Dump Schemes

Masterminds are entities organizing, coordinating, and orchestrating cryptocurrency pump-and-dump schemes, a form of trade-based manipulation undermining market integrity and causing financial losses for unwitting investors. Previous research detects pump-and-dump activities in the market, predicts the target cryptocurrency, and examines investors and \ac{osn} entities. However, these solutions do not address the root cause of the problem. There is a critical gap in identifying and tracing the masterminds involved in these schemes. In this research, we develop a detection system \textsc{Perseus}, which collects real-time data from the \acs{osn} and cryptocurrency markets. \textsc{Perseus} then constructs temporal attributed graphs that preserve the direction of information diffusion and the structure of the community while leveraging \ac{gnn} to identify the masterminds behind pump-and-dump activities. Our design of \textsc{Perseus} leads to higher F1 scores and precision than the \ac{sota} fraud detection method, achieving fast training and inferring speeds. Deployed in the real world from February 16 to October 9 2024, \textsc{Perseus} successfully detects $438$ masterminds who are efficient in the pump-and-dump information diffusion networks. \textsc{Perseus} provides regulators with an explanation of the risks of masterminds and oversight capabilities to mitigate the pump-and-dump schemes of cryptocurrency.

cs.CY↗

arXiv2025

Exploiting Vulnerabilities in Speech Translation Systems through Targeted Adversarial Attacks

As speech translation (ST) systems become increasingly prevalent, understanding their vulnerabilities is crucial for ensuring robust and reliable communication. However, limited work has explored this issue in depth. This paper explores methods of compromising these systems through imperceptible audio manipulations. Specifically, we present two innovative approaches: (1) the injection of perturbation into source audio, and (2) the generation of adversarial music designed to guide targeted translation, while also conducting more practical over-the-air attacks in the physical world. Our experiments reveal that carefully crafted audio perturbations can mislead translation models to produce targeted, harmful outputs, while adversarial music achieve this goal more covertly, exploiting the natural imperceptibility of music. These attacks prove effective across multiple languages and translation models, highlighting a systemic vulnerability in current ST architectures. The implications of this research extend beyond immediate security concerns, shedding light on the interpretability and robustness of neural speech processing systems. Our findings underscore the need for advanced defense mechanisms and more resilient architectures in the realm of audio systems. More details and samples can be found at https://adv-st.github.io.

cs.SD↗