arXiv subjects

Binghui Wang

Publications and source records attributed to Binghui Wang.

At least 19 recordsLinked to original sources

arXiv2026

Attention Sinks and Outliers in Attention Residuals

We propose OASIS, an outlier- and sink-aware technique built on inter-layer null signaling. As AttnResidual architectures introduce an additional depth-wise normalization channel, they improve inter-layer routing flexibility but also exacerbate attention sinks, activation outliers, and the resulting degradation in inference stability and quantization robustness. OASIS addresses this issue by introducing a Softmax1-based null space and coupling token-level null evidence to depth routing through an inter-layer null signal, thereby reducing sink-dominated routing and improving structural robustness. Theoretically, we show that the dual-normalization design of AttnResidual intensifies sink formation and quantization brittleness. Experimentally, we compare OASIS against five baselines on three real-world datasets and observe consistent improvements in both attention sink and post-quantization performance. Notably, OASIS achieves an average reduction of 9.26% in maximum infinity norm and 2.60% in average kurtosis across the evaluated settings, while lowering perplexity by 75.85% under W8A8 and improving GSM8K Pass@1 by 12.42% under W4A4.

cs.LG↗

arXiv2026

MoCo-EA: Exploiting Adversarial Mode Connectivity for Efficient Evolutionary Attacks

Evolutionary algorithms for adversarial attacks leverage population-based search to discover perturbations without gradient information, but suffer from inefficient crossover operations that destroy adversarial properties through discrete interpolation. We introduce Mode Connectivity Evolutionary Attack (MoCo-EA), which replaces traditional crossover with a novel B\'ezier crossover operator that optimizes perturbations along a continuous B\'ezier curve between parent perturbations. Our key insight is that adversarial examples lie on connected manifolds where intermediate points maintain and often enhance attack effectiveness. We demonstrate three findings: (1) Successful adversarial perturbations exhibit mode connectivity; (2) Intermediate points along optimized paths achieve higher transferability than endpoints; (3) B\'ezier crossover dramatically outperforms discrete genetic operations while reducing convergence time and query requirements. By exploiting the geometric structure of adversarial space through path optimization, MoCo-EA provides an efficient and reliable method. Our work challenges the traditional view of adversarial examples as isolated points and opens new directions for both attack generation and defense research.

cs.CR↗

arXiv2026

Contrastive Reasoning Alignment: Reinforcement Learning from Hidden Representations

We propose CRAFT, a red-teaming alignment framework that leverages model reasoning capabilities and hidden representations to improve robustness against jailbreak attacks. Unlike prior defenses that operate primarily at the output level, CRAFT aligns large reasoning models to generate safety-aware reasoning traces by explicitly optimizing objectives defined over the hidden state space. Methodologically, CRAFT integrates contrastive representation learning with reinforcement learning to separate safe and unsafe reasoning trajectories, yielding a latent-space geometry that supports robust, reasoning-level safety alignment. Theoretically, we show that incorporating latent-textual consistency into GRPO eliminates superficially aligned policies by ruling them out as local optima. Empirically, we evaluate CRAFT on multiple safety benchmarks using two strong reasoning models, Qwen3-4B-Thinking and R1-Distill-Llama-8B, where it consistently outperforms state-of-the-art defenses such as IPO and SafeKey. Notably, CRAFT delivers an average 79.0% improvement in reasoning safety and 87.7% improvement in final-response safety over the base models, demonstrating the effectiveness of hidden-space reasoning alignment.

cs.AI↗

arXiv2026

When One Modality Rules Them All: Backdoor Modality Collapse in Multimodal Diffusion Models

While diffusion models have revolutionized visual content generation, their rapid adoption has underscored the critical need to investigate vulnerabilities, e.g., to backdoor attacks. In multimodal diffusion models, it is natural to expect that attacking multiple modalities simultaneously (e.g., text and image) would yield complementary effects and strengthen the overall backdoor. In this paper, we challenge this assumption by investigating the phenomenon of Backdoor Modality Collapse, a scenario where the backdoor mechanism degenerates to rely predominantly on a subset of modalities, rendering others redundant. To rigorously quantify this behavior, we introduce two novel metrics: Trigger Modality Attribution (TMA) and Cross-Trigger Interaction (CTI). Through extensive experiments across diverse training configurations in multimodal conditional diffusion, we consistently observe a ``winner-takes-all'' dynamic in backdoor behavior. Our results reveal that (1) attacks often collapse into subset-modality dominance, and (2) cross-modal interaction is negligible or even negative, contradicting the intuition of synergistic vulnerability. These findings highlight a critical blind spot in current assessments, suggesting that high attack success rates often mask a fundamental reliance on a subset of modalities. This establishes a principled foundation for mechanistic analysis and future defense development.

cs.LG↗

arXiv2026

On Google's SynthID-Text LLM Watermarking System: Theoretical Analysis and Empirical Validation

Google's SynthID-Text, the first ever production-ready generative watermark system for large language model, designs a novel Tournament-based method that achieves the state-of-the-art detectability for identifying AI-generated texts. The system's innovation lies in: 1) a new Tournament sampling algorithm for watermarking embedding, 2) a detection strategy based on the introduced score function (e.g., Bayesian or mean score), and 3) a unified design that supports both distortionary and non-distortionary watermarking methods. This paper presents the first theoretical analysis of SynthID-Text, with a focus on its detection performance and watermark robustness, complemented by empirical validation. For example, we prove that the mean score is inherently vulnerable to increased tournament layers, and design a layer inflation attack to break SynthID-Text. We also prove the Bayesian score offers improved watermark robustness w.r.t. layers and further establish that the optimal Bernoulli distribution for watermark detection is achieved when the parameter is set to 0.5. Together, these theoretical and empirical insights not only deepen our understanding of SynthID-Text, but also open new avenues for analyzing effective watermark removal strategies and designing robust watermarking techniques. Source code is available at https: //github.com/romidi80/Synth-ID-Empirical-Analysis.

cs.CR↗

arXiv2026

SilentDrift: Exploiting Action Chunking for Stealthy Backdoor Attacks on Vision-Language-Action Models

Vision-Language-Action (VLA) models are increasingly deployed in safety-critical robotic applications, yet their security vulnerabilities remain underexplored. We identify a fundamental security flaw in modern VLA systems: the combination of action chunking and delta pose representations creates an intra-chunk visual open-loop. This mechanism forces the robot to execute K-step action sequences, allowing per-step perturbations to accumulate through integration. We propose SILENTDRIFT, a stealthy black-box backdoor attack exploiting this vulnerability. Our method employs the Smootherstep function to construct perturbations with guaranteed C2 continuity, ensuring zero velocity and acceleration at trajectory boundaries to satisfy strict kinematic consistency constraints. Furthermore, our keyframe attack strategy selectively poisons only the critical approach phase, maximizing impact while minimizing trigger exposure. The resulting poisoned trajectories are visually indistinguishable from successful demonstrations. Evaluated on the LIBERO, SILENTDRIFT achieves a 93.2% Attack Success Rate with a poisoning rate under 2%, while maintaining a 95.3% Clean Task Success Rate.

cs.CR↗

arXiv2025

Towards Strong Certified Defense with Universal Asymmetric Randomization

Randomized smoothing has become essential for achieving certified adversarial robustness in machine learning models. However, current methods primarily use isotropic noise distributions that are uniform across all data dimensions, such as image pixels, limiting the effectiveness of robustness certification by ignoring the heterogeneity of inputs and data dimensions. To address this limitation, we propose UCAN: a novel technique that \underline{U}niversally \underline{C}ertifies adversarial robustness with \underline{A}nisotropic \underline{N}oise. UCAN is designed to enhance any existing randomized smoothing method, transforming it from symmetric (isotropic) to asymmetric (anisotropic) noise distributions, thereby offering a more tailored defense against adversarial attacks. Our theoretical framework is versatile, supporting a wide array of noise distributions for certified robustness in different $\ell_p$-norms and applicable to any arbitrary classifier by guaranteeing the classifier's prediction over perturbed inputs with provable robustness bounds through tailored noise injection. Additionally, we develop a novel framework equipped with three exemplary noise parameter generators (NPGs) to optimally fine-tune the anisotropic noise parameters for different data dimensions, allowing for pursuing different levels of robustness enhancements in practice.Empirical evaluations underscore the significant leap in UCAN's performance over existing state-of-the-art methods, demonstrating up to $182.6\%$ improvement in certified accuracy at large certified radii on MNIST, CIFAR10, and ImageNet datasets.\footnote{Code is anonymously available at \href{https://github.com/youbin2014/UCAN/}{https://github.com/youbin2014/UCAN/}}

cs.LG↗

arXiv2025

Measure-Theoretic Anti-Causal Representation Learning

Causal representation learning in the anti-causal setting (labels cause features rather than the reverse) presents unique challenges requiring specialized approaches. We propose Anti-Causal Invariant Abstractions (ACIA), a novel measure-theoretic framework for anti-causal representation learning. ACIA employs a two-level design, low-level representations capture how labels generate observations, while high-level representations learn stable causal patterns across environment-specific variations. ACIA addresses key limitations of existing approaches by accommodating prefect and imperfect interventions through interventional kernels, eliminating dependency on explicit causal structures, handling high-dimensional data effectively, and providing theoretical guarantees for out-of-distribution generalization. Experiments on synthetic and real-world medical datasets demonstrate that ACIA consistently outperforms state-of-the-art methods in both accuracy and invariance metrics. Furthermore, our theoretical results establish tight bounds on performance gaps between training and unseen environments, confirming the efficacy of our approach for robust anti-causal learning.

cs.LG↗

arXiv2025

Practical, Generalizable and Robust Backdoor Attacks on Text-to-Image Diffusion Models

Text-to-image diffusion models (T2I DMs) have achieved remarkable success in generating high-quality and diverse images from text prompts, yet recent studies have revealed their vulnerability to backdoor attacks. Existing attack methods suffer from critical limitations: 1) they rely on unnatural adversarial prompts that lack human readability and require massive poisoned data; 2) their effectiveness is typically restricted to specific models, lacking generalizability; and 3) they can be mitigated by recent backdoor defenses. To overcome these challenges, we propose a novel backdoor attack framework that achieves three key properties: 1) \emph{Practicality}: Our attack requires only a few stealthy backdoor samples to generate arbitrary attacker-chosen target images, as well as ensuring high-quality image generation in benign scenarios. 2) \emph{Generalizability:} The attack is applicable across multiple T2I DMs without requiring model-specific redesign. 3) \emph{Robustness:} The attack remains effective against existing backdoor defenses and adaptive defenses. Our extensive experimental results on multiple T2I DMs demonstrate that with only 10 carefully crafted backdoored samples, our attack method achieves $>$90\% attack success rate with negligible degradation in benign image generation quality. We also conduct human evaluation to validate our attack effectiveness. Furthermore, recent backdoor detection and mitigation methods, as well as adaptive defense tailored to our attack are not sufficiently effective, highlighting the pressing need for more robust defense mechanisms against the proposed attack.

cs.CR↗

arXiv2025

On the Inference (In-)Security of Vertical Federated Learning: Efficient Auditing against Inference Tampering Attack

Vertical Federated Learning (VFL) is an emerging distributed learning paradigm for cross-silo collaboration without accessing participants' data. However, existing VFL work lacks a mechanism to audit the inference correctness of the data party. The malicious data party can modify the local data and model to mislead the joint inference results. To exploit this vulnerability, we design a novel Vertical Federated Inference Tampering (VeFIT) attack, allowing the data party to covertly tamper with the local inference and mislead results on the task party's final prediction. VeFIT can decrease the task party's inference accuracy by an average of 34.49%. Existing defense mechanisms can not effectively detect this attack, and the detection performance is near random guessing. To mitigate the attack, we further design a Vertical Federated Inference Auditing (VeFIA) framework. VeFIA helps the task party to audit whether the data party's inferences are executed as expected during large-scale online inference. VeFIA does not leak the data party's privacy nor introduce additional latency. The core design is that the task party can use the inference results from a framework with Trusted Execution Environments (TEE) and the coordinator to validate the correctness of the data party's computation results. VeFIA guarantees that, as long as the proportion of inferences attacked by VeFIT exceeds 5.4%, the task party can detect the malicious behavior of the data party with a probability of 99.99%, without any additional online overhead. VeFIA's random sampling validation of VeFIA achieves 100% positive predictive value, negative predictive value, and true positive rate in detecting VeFIT. We further validate VeFIA's effectiveness in terms of privacy protection and scalability on real-world datasets. To the best of our knowledge, this is the first paper discussing the inference auditing problem towards VFL.

cs.SE↗

arXiv2025

Rectifying Privacy and Efficacy Measurements in Machine Unlearning: A New Inference Attack Perspective

Machine unlearning focuses on efficiently removing specific data from trained models, addressing privacy and compliance concerns with reasonable costs. Although exact unlearning ensures complete data removal equivalent to retraining, it is impractical for large-scale models, leading to growing interest in inexact unlearning methods. However, the lack of formal guarantees in these methods necessitates the need for robust evaluation frameworks to assess their privacy and effectiveness. In this work, we first identify several key pitfalls of the existing unlearning evaluation frameworks, e.g., focusing on average-case evaluation or targeting random samples for evaluation, incomplete comparisons with the retraining baseline. Then, we propose RULI (Rectified Unlearning Evaluation Framework via Likelihood Inference), a novel framework to address critical gaps in the evaluation of inexact unlearning methods. RULI introduces a dual-objective attack to measure both unlearning efficacy and privacy risks at a per-sample granularity. Our findings reveal significant vulnerabilities in state-of-the-art unlearning methods, where RULI achieves higher attack success rates, exposing privacy risks underestimated by existing methods. Built on a game-based foundation and validated through empirical evaluations on both image and text data (spanning tasks from classification to generation), RULI provides a rigorous, scalable, and fine-grained methodology for evaluating unlearning techniques.

cs.CR↗

arXiv2025

GenoArmory: A Unified Evaluation Framework for Adversarial Attacks on Genomic Foundation Models

We propose the first unified adversarial attack benchmark for Genomic Foundation Models (GFMs), named GenoArmory. Unlike existing GFM benchmarks, GenoArmory offers the first comprehensive evaluation framework to systematically assess the vulnerability of GFMs to adversarial attacks. Methodologically, we evaluate the adversarial robustness of five state-of-the-art GFMs using four widely adopted attack algorithms and three defense strategies. Importantly, our benchmark provides an accessible and comprehensive framework to analyze GFM vulnerabilities with respect to model architecture, quantization schemes, and training datasets. Additionally, we introduce GenoAdv, a new adversarial sample dataset designed to improve GFM safety. Empirically, classification models exhibit greater robustness to adversarial perturbations compared to generative models, highlighting the impact of task type on model vulnerability. Moreover, adversarial attacks frequently target biologically significant genomic regions, suggesting that these models effectively capture meaningful sequence features.

cs.LG↗

arXiv2025

Deterministic Certification of Graph Neural Networks against Graph Poisoning Attacks with Arbitrary Perturbations

Graph neural networks (GNNs) are becoming the de facto method to learn on the graph data and have achieved the state-of-the-art on node and graph classification tasks. However, recent works show GNNs are vulnerable to training-time poisoning attacks -- marginally perturbing edges, nodes, or/and node features of training graph(s) can largely degrade GNNs' testing performance. Most previous defenses against graph poisoning attacks are empirical and are soon broken by adaptive / stronger ones. A few provable defenses provide robustness guarantees, but have large gaps when applied in practice: 1) restrict the attacker on only one type of perturbation; 2) design for a particular GNN architecture or task; and 3) robustness guarantees are not 100\% accurate. In this work, we bridge all these gaps by developing PGNNCert, the first certified defense of GNNs against poisoning attacks under arbitrary (edge, node, and node feature) perturbations with deterministic robustness guarantees. Extensive evaluations on multiple node and graph classification datasets and GNNs demonstrate the effectiveness of PGNNCert to provably defend against arbitrary poisoning perturbations. PGNNCert is also shown to significantly outperform the state-of-the-art certified defenses against edge perturbation or node perturbation during GNN training.

cs.LG↗

arXiv2025

FedTilt: Towards Multi-Level Fairness-Preserving and Robust Federated Learning

Federated Learning (FL) is an emerging decentralized learning paradigm that can partly address the privacy concern that cannot be handled by traditional centralized and distributed learning. Further, to make FL practical, it is also necessary to consider constraints such as fairness and robustness. However, existing robust FL methods often produce unfair models, and existing fair FL methods only consider one-level (client) fairness and are not robust to persistent outliers (i.e., injected outliers into each training round) that are common in real-world FL settings. We propose \texttt{FedTilt}, a novel FL that can preserve multi-level fairness and be robust to outliers. In particular, we consider two common levels of fairness, i.e., \emph{client fairness} -- uniformity of performance across clients, and \emph{client data fairness} -- uniformity of performance across different classes of data within a client. \texttt{FedTilt} is inspired by the recently proposed tilted empirical risk minimization, which introduces tilt hyperparameters that can be flexibly tuned. Theoretically, we show how tuning tilt values can achieve the two-level fairness and mitigate the persistent outliers, and derive the convergence condition of \texttt{FedTilt} as well. Empirically, our evaluation results on a suite of realistic federated datasets in diverse settings show the effectiveness and flexibility of the \texttt{FedTilt} framework and the superiority to the state-of-the-arts.

cs.LG↗

arXiv2025

Backdoor Attacks on Discrete Graph Diffusion Models

Diffusion models are powerful generative models in continuous data domains such as image and video data. Discrete graph diffusion models (DGDMs) have recently extended them for graph generation, which are crucial in fields like molecule and protein modeling, and obtained the SOTA performance. However, it is risky to deploy DGDMs for safety-critical applications (e.g., drug discovery) without understanding their security vulnerabilities. In this work, we perform the first study on graph diffusion models against backdoor attacks, a severe attack that manipulates both the training and inference/generation phases in graph diffusion models. We first define the threat model, under which we design the attack such that the backdoored graph diffusion model can generate 1) high-quality graphs without backdoor activation, 2) effective, stealthy, and persistent backdoored graphs with backdoor activation, and 3) graphs that are permutation invariant and exchangeable--two core properties in graph generative models. 1) and 2) are validated via empirical evaluations without and with backdoor defenses, while 3) is validated via theoretical results.

cs.CR↗

arXiv2025

Provably Robust Explainable Graph Neural Networks against Graph Perturbation Attacks

Explaining Graph Neural Network (XGNN) has gained growing attention to facilitate the trust of using GNNs, which is the mainstream method to learn graph data. Despite their growing attention, Existing XGNNs focus on improving the explanation performance, and its robustness under attacks is largely unexplored. We noticed that an adversary can slightly perturb the graph structure such that the explanation result of XGNNs is largely changed. Such vulnerability of XGNNs could cause serious issues particularly in safety/security-critical applications. In this paper, we take the first step to study the robustness of XGNN against graph perturbation attacks, and propose XGNNCert, the first provably robust XGNN. Particularly, our XGNNCert can provably ensure the explanation result for a graph under the worst-case graph perturbation attack is close to that without the attack, while not affecting the GNN prediction, when the number of perturbed edges is bounded. Evaluation results on multiple graph datasets and GNN explainers show the effectiveness of XGNNCert.

cs.CR↗

arXiv2025

AGNNCert: Defending Graph Neural Networks against Arbitrary Perturbations with Deterministic Certification

Graph neural networks (GNNs) achieve the state-of-the-art on graph-relevant tasks such as node and graph classification. However, recent works show GNNs are vulnerable to adversarial perturbations include the perturbation on edges, nodes, and node features, the three components forming a graph. Empirical defenses against such attacks are soon broken by adaptive ones. While certified defenses offer robustness guarantees, they face several limitations: 1) almost all restrict the adversary's capability to only one type of perturbation, which is impractical; 2) all are designed for a particular GNN task, which limits their applicability; and 3) the robustness guarantees of all methods except one are not 100% accurate. We address all these limitations by developing AGNNCert, the first certified defense for GNNs against arbitrary (edge, node, and node feature) perturbations with deterministic robustness guarantees, and applicable to the two most common node and graph classification tasks. AGNNCert also encompass existing certified defenses as special cases. Extensive evaluations on multiple benchmark node/graph classification datasets and two real-world graph datasets, and multiple GNNs validate the effectiveness of AGNNCert to provably defend against arbitrary perturbations. AGNNCert also shows its superiority over the state-of-the-art certified defenses against the individual edge perturbation and node perturbation.

cs.CR↗

arXiv2025

Watermarking Graph Neural Networks via Explanations for Ownership Protection

Graph Neural Networks (GNNs) are widely deployed in industry, making their intellectual property valuable. However, protecting GNNs from unauthorized use remains a challenge. Watermarking offers a solution by embedding ownership information into models. Existing watermarking methods have two limitations: First, they rarely focus on graph data or GNNs. Second, the de facto backdoor-based method relies on manipulating training data, which can introduce ownership ambiguity through misclassification and vulnerability to data poisoning attacks that can interrupt the backdoor mechanism. Our explanation-based watermarking inherits the strengths of backdoor-based methods (e.g., black-box verification) without data manipulation, eliminating ownership ambiguity and data dependencies. In particular, we watermark GNN explanations such that these explanations are statistically distinct from others, so ownership claims must be verified through statistical significance. We theoretically prove that, even with full knowledge of our method, locating the watermark is NP-hard. Empirically, our method demonstrates robustness to fine-tuning and pruning attacks. By addressing these challenges, our approach significantly advances GNN intellectual property protection.

cs.CR↗