arXiv subjects

Boheng Li

Publications and source records attributed to Boheng Li.

At least 19 recordsLinked to original sources

arXiv2026

Mitigating Many-shot Jailbreak Attacks with One Single Demonstration

Many-shot jailbreaking (MSJ) causes safety-aligned language models to answer harmful queries by preceding them with many harmful question-answer demonstrations. We study why this attack becomes stronger as the number of demonstrations increases. Empirically, we find that MSJ induces a progressive activation drift: the representation of a fixed harmful query moves step by step away from the safety-aligned region as more harmful demonstrations are added. Theoretically, we show that this drift can be interpreted as implicit malicious fine-tuning: conditioning on N harmful demonstrations induces SGD-style updates equivalent to optimizing on the corresponding N harmful samples. This view turns the attack mechanism into a defense principle. We append a fixed one-shot safety demonstration at inference time, which induces a counteracting safety-oriented update and restores refusal behavior. The resulting method improves the model's robustness to MSJ without modifying its parameters or requiring white-box access at deployment. Code is available at https://github.com/Thecommonirin/SafeEnd.

cs.CR↗

arXiv2025

External Data Extraction Attacks against Retrieval-Augmented Large Language Models

In recent years, RAG has emerged as a key paradigm for enhancing large language models (LLMs). By integrating externally retrieved information, RAG alleviates issues like outdated knowledge and, crucially, insufficient domain expertise. While effective, RAG introduces new risks of external data extraction attacks (EDEAs), where sensitive or copyrighted data in its knowledge base may be extracted verbatim. These risks are particularly acute when RAG is used to customize specialized LLM applications with private knowledge bases. Despite initial studies exploring these risks, they often lack a formalized framework, robust attack performance, and comprehensive evaluation, leaving critical questions about real-world EDEA feasibility unanswered. In this paper, we present the first comprehensive study to formalize EDEAs against retrieval-augmented LLMs. We first formally define EDEAs and propose a unified framework decomposing their design into three components: extraction instruction, jailbreak operator, and retrieval trigger, under which prior attacks can be considered instances within our framework. Guided by this framework, we develop SECRET: a Scalable and EffeCtive exteRnal data Extraction aTtack. Specifically, SECRET incorporates (1) an adaptive optimization process using LLMs as optimizers to generate specialized jailbreak prompts for EDEAs, and (2) cluster-focused triggering, an adaptive strategy that alternates between global exploration and local exploitation to efficiently generate effective retrieval triggers. Extensive evaluations across 4 models reveal that SECRET significantly outperforms previous attacks, and is highly effective against all 16 tested RAG instances. Notably, SECRET successfully extracts 35% of the data from RAG powered by Claude 3.7 Sonnet for the first time, whereas other attacks yield 0% extraction. Our findings call for attention to this emerging threat.

cs.CR↗

arXiv2025

Taught Well Learned Ill: Towards Distillation-conditional Backdoor Attack

Knowledge distillation (KD) is a vital technique for deploying deep neural networks (DNNs) on resource-constrained devices by transferring knowledge from large teacher models to lightweight student models. While teacher models from third-party platforms may undergo security verification (\eg, backdoor detection), we uncover a novel and critical threat: distillation-conditional backdoor attacks (DCBAs). DCBA injects dormant and undetectable backdoors into teacher models, which become activated in student models via the KD process, even with clean distillation datasets. While the direct extension of existing methods is ineffective for DCBA, we implement this attack by formulating it as a bilevel optimization problem and proposing a simple yet effective method (\ie, SCAR). Specifically, the inner optimization simulates the KD process by optimizing a surrogate student model, while the outer optimization leverages outputs from this surrogate to optimize the teacher model for implanting the conditional backdoor. Our SCAR addresses this complex optimization utilizing an implicit differentiation algorithm with a pre-optimized trigger injection function. Extensive experiments across diverse datasets, model architectures, and KD techniques validate the effectiveness of our SCAR and its resistance against existing backdoor detection, highlighting a significant yet previously overlooked vulnerability in the KD process. Our code is available at https://github.com/WhitolfChen/SCAR.

cs.CR↗

arXiv2025

Towards Resilient Safety-driven Unlearning for Diffusion Models against Downstream Fine-tuning

Text-to-image (T2I) diffusion models have achieved impressive image generation quality and are increasingly fine-tuned for personalized applications. However, these models often inherit unsafe behaviors from toxic pretraining data, raising growing safety concerns. While recent safety-driven unlearning methods have made promising progress in suppressing model toxicity, they are found to be fragile to downstream fine-tuning, as we reveal that state-of-the-art methods largely fail to retain their effectiveness even when fine-tuned on entirely benign datasets. To mitigate this problem, in this paper, we propose ResAlign, a safety-driven unlearning framework with enhanced resilience against downstream fine-tuning. By modeling downstream fine-tuning as an implicit optimization problem with a Moreau envelope-based reformulation, ResAlign enables efficient gradient estimation to minimize the recovery of harmful behaviors. Additionally, a meta-learning strategy is proposed to simulate a diverse distribution of fine-tuning scenarios to improve generalization. Extensive experiments across a wide range of datasets, fine-tuning methods, and configurations demonstrate that ResAlign consistently outperforms prior unlearning approaches in retaining safety, while effectively preserving benign generation capability. Our code and pretrained models are publicly available at https://github.com/AntigoneRandy/ResAlign.

cs.LG↗

arXiv2025

DREAM: Scalable Red Teaming for Text-to-Image Generative Systems via Distribution Modeling

Despite the integration of safety alignment and external filters, text-to-image (T2I) generative systems are still susceptible to producing harmful content, such as sexual or violent imagery. This raises serious concerns about unintended exposure and potential misuse. Red teaming, which aims to proactively identify diverse prompts that can elicit unsafe outputs from the T2I system, is increasingly recognized as an essential method for assessing and improving safety before real-world deployment. However, existing automated red teaming approaches often treat prompt discovery as an isolated, prompt-level optimization task, which limits their scalability, diversity, and overall effectiveness. To bridge this gap, in this paper, we propose DREAM, a scalable red teaming framework to automatically uncover diverse problematic prompts from a given T2I system. Unlike prior work that optimizes prompts individually, DREAM directly models the probabilistic distribution of the target system's problematic prompts, which enables explicit optimization over both effectiveness and diversity, and allows efficient large-scale sampling after training. To achieve this without direct access to representative training samples, we draw inspiration from energy-based models and reformulate the objective into a simple and tractable form. We further introduce GC-SPSA, an efficient optimization algorithm that provides stable gradient estimates through the long and potentially non-differentiable T2I pipeline. During inference, we also propose a diversity-aware sampling strategy to enhance prompt variety. The effectiveness of DREAM is validated through extensive experiments, demonstrating state-of-the-art performance across a wide range of T2I models and safety filters in terms of both prompt success rate and diversity. Our code is available at https://github.com/AntigoneRandy/DREAM

cs.CR↗

arXiv2025

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 in https://github.com/shaoshuo-ss/DATABench.

cs.CR↗

arXiv2025

Revisiting Data Auditing in Large Vision-Language Models

With the surge of large language models (LLMs), Large Vision-Language Models (VLMs)--which integrate vision encoders with LLMs for accurate visual grounding--have shown great potential in tasks like generalist agents and robotic control. However, VLMs are typically trained on massive web-scraped images, raising concerns over copyright infringement and privacy violations, and making data auditing increasingly urgent. Membership inference (MI), which determines whether a sample was used in training, has emerged as a key auditing technique, with promising results on open-source VLMs like LLaVA (AUC > 80%). In this work, we revisit these advances and uncover a critical issue: current MI benchmarks suffer from distribution shifts between member and non-member images, introducing shortcut cues that inflate MI performance. We further analyze the nature of these shifts and propose a principled metric based on optimal transport to quantify the distribution discrepancy. To evaluate MI in realistic settings, we construct new benchmarks with i.i.d. member and non-member images. Existing MI methods fail under these unbiased conditions, performing only marginally better than chance. Further, we explore the theoretical upper bound of MI by probing the Bayes Optimality within the VLM's embedding space and find the irreducible error rate remains high. Despite this pessimistic outlook, we analyze why MI for VLMs is particularly challenging and identify three practical scenarios--fine-tuning, access to ground-truth texts, and set-based inference--where auditing becomes feasible. Our study presents a systematic view of the limits and opportunities of MI for VLMs, providing guidance for future efforts in trustworthy data auditing.

cs.CV↗

arXiv2025

Towards Label-Only Membership Inference Attack against Pre-trained Large Language Models

Membership Inference Attacks (MIAs) aim to predict whether a data sample belongs to the model's training set or not. Although prior research has extensively explored MIAs in Large Language Models (LLMs), they typically require accessing to complete output logits (\ie, \textit{logits-based attacks}), which are usually not available in practice. In this paper, we study the vulnerability of pre-trained LLMs to MIAs in the \textit{label-only setting}, where the adversary can only access generated tokens (text). We first reveal that existing label-only MIAs have minor effects in attacking pre-trained LLMs, although they are highly effective in inferring fine-tuning datasets used for personalized LLMs. We find that their failure stems from two main reasons, including better generalization and overly coarse perturbation. Specifically, due to the extensive pre-training corpora and exposing each sample only a few times, LLMs exhibit minimal robustness differences between members and non-members. This makes token-level perturbations too coarse to capture such differences. To alleviate these problems, we propose \textbf{PETAL}: a label-only membership inference attack based on \textbf{PE}r-\textbf{T}oken sem\textbf{A}ntic simi\textbf{L}arity. Specifically, PETAL leverages token-level semantic similarity to approximate output probabilities and subsequently calculate the perplexity. It finally exposes membership based on the common assumption that members are `better' memorized and have smaller perplexity. We conduct extensive experiments on the WikiMIA benchmark and the more challenging MIMIR benchmark. Empirically, our PETAL performs better than the extensions of existing label-only attacks against personalized LLMs and even on par with other advanced logit-based attacks across all metrics on five prevalent open-source LLMs.

cs.CR↗

arXiv2024

Towards Reliable Verification of Unauthorized Data Usage in Personalized Text-to-Image Diffusion Models

Text-to-image diffusion models are pushing the boundaries of what generative AI can achieve in our lives. Beyond their ability to generate general images, new personalization techniques have been proposed to customize the pre-trained base models for crafting images with specific themes or styles. Such a lightweight solution, enabling AI practitioners and developers to easily build their own personalized models, also poses a new concern regarding whether the personalized models are trained from unauthorized data. A promising solution is to proactively enable data traceability in generative models, where data owners embed external coatings (e.g., image watermarks or backdoor triggers) onto the datasets before releasing. Later the models trained over such datasets will also learn the coatings and unconsciously reproduce them in the generated mimicries, which can be extracted and used as the data usage evidence. However, we identify the existing coatings cannot be effectively learned in personalization tasks, making the corresponding verification less reliable. In this paper, we introduce SIREN, a novel methodology to proactively trace unauthorized data usage in black-box personalized text-to-image diffusion models. Our approach optimizes the coating in a delicate way to be recognized by the model as a feature relevant to the personalization task, thus significantly improving its learnability. We also utilize a human perceptual-aware constraint, a hypersphere classification technique, and a hypothesis-testing-guided verification method to enhance the stealthiness and detection accuracy of the coating. The effectiveness of SIREN is verified through extensive experiments on a diverse set of benchmark datasets, models, and learning algorithms. SIREN is also effective in various real-world scenarios and evaluated against potential countermeasures. Our code is publicly available.

cs.CY↗

arXiv2024

Is Difficulty Calibration All We Need? Towards More Practical Membership Inference Attacks

The vulnerability of machine learning models to Membership Inference Attacks (MIAs) has garnered considerable attention in recent years. These attacks determine whether a data sample belongs to the model's training set or not. Recent research has focused on reference-based attacks, which leverage difficulty calibration with independently trained reference models. While empirical studies have demonstrated its effectiveness, there is a notable gap in our understanding of the circumstances under which it succeeds or fails. In this paper, we take a further step towards a deeper understanding of the role of difficulty calibration. Our observations reveal inherent limitations in calibration methods, leading to the misclassification of non-members and suboptimal performance, particularly on high-loss samples. We further identify that these errors stem from an imperfect sampling of the potential distribution and a strong dependence of membership scores on the model parameters. By shedding light on these issues, we propose RAPID: a query-efficient and computation-efficient MIA that directly \textbf{R}e-lever\textbf{A}ges the original membershi\textbf{P} scores to m\textbf{I}tigate the errors in \textbf{D}ifficulty calibration. Our experimental results, spanning 9 datasets and 5 model architectures, demonstrate that RAPID outperforms previous state-of-the-art attacks (e.g., LiRA and Canary offline) across different metrics while remaining computationally efficient. Our observations and analysis challenge the current de facto paradigm of difficulty calibration in high-precision inference, encouraging greater attention to the persistent risks posed by MIAs in more practical scenarios.

cs.CR↗

arXiv2024

Invisible Backdoor Attack against Self-supervised Learning

Self-supervised learning (SSL) models are vulnerable to backdoor attacks. Existing backdoor attacks that are effective in SSL often involve noticeable triggers, like colored patches or visible noise, which are vulnerable to human inspection. This paper proposes an imperceptible and effective backdoor attack against self-supervised models. We first find that existing imperceptible triggers designed for supervised learning are less effective in compromising self-supervised models. We then identify this ineffectiveness is attributed to the overlap in distributions between the backdoor and augmented samples used in SSL. Building on this insight, we design an attack using optimized triggers disentangled with the augmented transformation in the SSL, while remaining imperceptible to human vision. Experiments on five datasets and six SSL algorithms demonstrate our attack is highly effective and stealthy. It also has strong resistance to existing backdoor defenses. Our code can be found at https://github.com/Zhang-Henry/INACTIVE.

cs.CV↗

arXiv2024

Nearest is Not Dearest: Towards Practical Defense against Quantization-conditioned Backdoor Attacks

Model quantization is widely used to compress and accelerate deep neural networks. However, recent studies have revealed the feasibility of weaponizing model quantization via implanting quantization-conditioned backdoors (QCBs). These special backdoors stay dormant on released full-precision models but will come into effect after standard quantization. Due to the peculiarity of QCBs, existing defenses have minor effects on reducing their threats or are even infeasible. In this paper, we conduct the first in-depth analysis of QCBs. We reveal that the activation of existing QCBs primarily stems from the nearest rounding operation and is closely related to the norms of neuron-wise truncation errors (i.e., the difference between the continuous full-precision weights and its quantized version). Motivated by these insights, we propose Error-guided Flipped Rounding with Activation Preservation (EFRAP), an effective and practical defense against QCBs. Specifically, EFRAP learns a non-nearest rounding strategy with neuron-wise error norm and layer-wise activation preservation guidance, flipping the rounding strategies of neurons crucial for backdoor effects but with minimal impact on clean accuracy. Extensive evaluations on benchmark datasets demonstrate that our EFRAP can defeat state-of-the-art QCB attacks under various settings. Code is available at https://github.com/AntigoneRandy/QuantBackdoor_EFRAP.

cs.CR↗

arXiv2024

Transfer Learning of Real Image Features with Soft Contrastive Loss for Fake Image Detection

In the last few years, the artifact patterns in fake images synthesized by different generative models have been inconsistent, leading to the failure of previous research that relied on spotting subtle differences between real and fake. In our preliminary experiments, we find that the artifacts in fake images always change with the development of the generative model, while natural images exhibit stable statistical properties. In this paper, we employ natural traces shared only by real images as an additional target for a classifier. Specifically, we introduce a self-supervised feature mapping process for natural trace extraction and develop a transfer learning based on soft contrastive loss to bring them closer to real images and further away from fake ones. This motivates the detector to make decisions based on the proximity of images to the natural traces. To conduct a comprehensive experiment, we built a high-quality and diverse dataset that includes generative models comprising GANs and diffusion models, to evaluate the effectiveness in generalizing unknown forgery techniques and robustness in surviving different transformations. Experimental results show that our proposed method gives 96.2% mAP significantly outperforms the baselines. Extensive experiments conducted on popular commercial platforms reveal that our proposed method achieves an accuracy exceeding 78.4%, underscoring its practicality for real-world application deployment.

cs.CV↗

arXiv2024

WHU-Synthetic: A Synthetic Perception Dataset for 3-D Multitask Model Research

End-to-end models capable of handling multiple sub-tasks in parallel have become a new trend, thereby presenting significant challenges and opportunities for the integration of multiple tasks within the domain of 3D vision. The limitations of 3D data acquisition conditions have not only restricted the exploration of many innovative research problems but have also caused existing 3D datasets to predominantly focus on single tasks. This has resulted in a lack of systematic approaches and theoretical frameworks for 3D multi-task learning, with most efforts merely serving as auxiliary support to the primary task. In this paper, we introduce WHU-Synthetic, a large-scale 3D synthetic perception dataset designed for multi-task learning, from the initial data augmentation (upsampling and depth completion), through scene understanding (segmentation), to macro-level tasks (place recognition and 3D reconstruction). Collected in the same environmental domain, we ensure inherent alignment across sub-tasks to construct multi-task models without separate training methods. Besides, we implement several novel settings, making it possible to realize certain ideas that are difficult to achieve in real-world scenarios. This supports more adaptive and robust multi-task perception tasks, such as sampling on city-level models, providing point clouds with different densities, and simulating temporal changes. Using our dataset, we conduct several experiments to investigate mutual benefits between sub-tasks, revealing new observations, challenges, and opportunities for future research. The dataset is accessible at https://github.com/WHU-USI3DV/WHU-Synthetic.

cs.CV↗

arXiv2024

Lips Are Lying: Spotting the Temporal Inconsistency between Audio and Visual in Lip-Syncing DeepFakes

In recent years, DeepFake technology has achieved unprecedented success in high-quality video synthesis, but these methods also pose potential and severe security threats to humanity. DeepFake can be bifurcated into entertainment applications like face swapping and illicit uses such as lip-syncing fraud. However, lip-forgery videos, which neither change identity nor have discernible visual artifacts, present a formidable challenge to existing DeepFake detection methods. Our preliminary experiments have shown that the effectiveness of the existing methods often drastically decrease or even fail when tackling lip-syncing videos. In this paper, for the first time, we propose a novel approach dedicated to lip-forgery identification that exploits the inconsistency between lip movements and audio signals. We also mimic human natural cognition by capturing subtle biological links between lips and head regions to boost accuracy. To better illustrate the effectiveness and advances of our proposed method, we create a high-quality LipSync dataset, AVLips, by employing the state-of-the-art lip generators. We hope this high-quality and diverse dataset could be well served the further research on this challenging and interesting field. Experimental results show that our approach gives an average accuracy of more than 95.3% in spotting lip-syncing videos, significantly outperforming the baselines. Extensive experiments demonstrate the capability to tackle deepfakes and the robustness in surviving diverse input transformations. Our method achieves an accuracy of up to 90.2% in real-world scenarios (e.g., WeChat video call) and shows its powerful capabilities in real scenario deployment. To facilitate the progress of this research community, we release all resources at https://github.com/AaronComo/LipFD.

cs.CV↗

arXiv2023

Dual-level Interaction for Domain Adaptive Semantic Segmentation

Self-training approach recently secures its position in domain adaptive semantic segmentation, where a model is trained with target domain pseudo-labels. Current advances have mitigated noisy pseudo-labels resulting from the domain gap. However, they still struggle with erroneous pseudo-labels near the boundaries of the semantic classifier. In this paper, we tackle this issue by proposing a dual-level interaction for domain adaptation (DIDA) in semantic segmentation. Explicitly, we encourage the different augmented views of the same pixel to have not only similar class prediction (semantic-level) but also akin similarity relationship with respect to other pixels (instance-level). As it's impossible to keep features of all pixel instances for a dataset, we, therefore, maintain a labeled instance bank with dynamic updating strategies to selectively store the informative features of instances. Further, DIDA performs cross-level interaction with scattering and gathering techniques to regenerate more reliable pseudo-labels. Our method outperforms the state-of-the-art by a notable margin, especially on confusing and long-tailed classes. Code is available at \href{https://github.com/RainJamesY/DIDA}

cs.CV↗

arXiv2023

What can Discriminator do? Towards Box-free Ownership Verification of Generative Adversarial Network

In recent decades, Generative Adversarial Network (GAN) and its variants have achieved unprecedented success in image synthesis. However, well-trained GANs are under the threat of illegal steal or leakage. The prior studies on remote ownership verification assume a black-box setting where the defender can query the suspicious model with specific inputs, which we identify is not enough for generation tasks. To this end, in this paper, we propose a novel IP protection scheme for GANs where ownership verification can be done by checking outputs only, without choosing the inputs (i.e., box-free setting). Specifically, we make use of the unexploited potential of the discriminator to learn a hypersphere that captures the unique distribution learned by the paired generator. Extensive evaluations on two popular GAN tasks and more than 10 GAN architectures demonstrate our proposed scheme to effectively verify the ownership. Our proposed scheme shown to be immune to popular input-based removal attacks and robust against other existing attacks. The source code and models are available at https://github.com/AbstractTeen/gan_ownership_verification

cs.CV↗

arXiv2022

Comprehensive Evaluation of Emergency Shelters in Wuhan City Based on GIS

Emergency shelters, which reflect the city's ability to respond to and deal with major public emergencies to a certain extent, are essential to a modern urban emergency management system. This paper is based on spatial analysis methods, using Analytic Hierarchy Process to analyze the suitability of the 28 emergency shelters in Wuhan City. The Technique for Order Preference by Similarity to an Ideal Solution is further used to evaluate the accommodation capacity of emergency shelters in central urban areas, which provides a reference for the optimization of existing shelters and the site selection of new shelters, and provides a basis for improving the service capacity of shelters. The results show that the overall situation of emergency shelters in Wuhan is good, with 96\% of the places reaching the medium level or above, but the suitability level needs to be further improved, especially the effectiveness and accessibility. Among the seven central urban areas in Wuhan, Hongshan District has the strongest accommodation capacity while Jianghan District has the weakest, with noticeable differences.

cs.CY↗