Abstract:Vision Transformers (ViTs) have been widely applied in various computer vision and vision-language tasks. To gain insights into their robustness in practical scenarios, transferable adversarial examples on ViTs have been extensively studied. A typical approach to improving adversarial transferability is by refining the surrogate model. However, existing work on ViTs has restricted their surrogate refinement to backward propagation. In this work, we instead focus on Forward Propagation Refinement (FPR) and specifically refine two key modules of ViTs: attention maps and token embeddings. For attention maps, we propose Attention Map Diversification (AMD), which diversifies certain attention maps and also implicitly imposes beneficial gradient vanishing during backward propagation. For token embeddings, we propose Momentum Token Embedding (MTE), which accumulates historical token embeddings to stabilize the forward updates in both the Attention and MLP blocks. We conduct extensive experiments with adversarial examples transferred from ViTs to various CNNs and ViTs, demonstrating that our FPR outperforms the current best (backward) surrogate refinement by up to 7.0\% on average. We also validate its superiority against popular defenses and its compatibility with other transfer methods. Codes and appendix are available at https://github.com/RYC-98/FPR.
Abstract:Backdoor attacks typically place a specific trigger on certain training data, such that the model makes prediction errors on inputs with that trigger during inference. Despite the core role of the trigger, existing studies have commonly believed a perfect match between training-inference triggers is optimal. In this paper, for the first time, we systematically explore the training-inference trigger relation, particularly focusing on their mismatch, based on a Training-Inference Trigger Intensity Manipulation (TITIM) workflow. TITIM specifically investigates the training-inference trigger intensity, such as the size or the opacity of a trigger, and reveals new insights into trigger generalization and overfitting. These new insights challenge the above common belief by demonstrating that the training-inference trigger mismatch can facilitate attacks in two practical scenarios, posing more significant security threats than previously thought. First, when the inference trigger is fixed, using training triggers with mixed intensities leads to stronger attacks than using any single intensity. For example, on CIFAR-10 with ResNet-18, mixing training triggers with 1.0 and 0.1 opacities improves the worst-case attack success rate (ASR) (over different testing opacities) of the best single-opacity attack from 10.61\% to 92.77\%. Second, intentionally using certain mismatched training-inference triggers can improve the attack stealthiness, i.e., better bypassing defenses. For example, compared to the training/inference intensity of 1.0/1.0, using 1.0/0.7 decreases the area under the curve (AUC) of the Scale-Up defense from 0.96 to 0.62, while maintaining a high attack ASR (99.65\% vs. 91.62\%). The above new insights are validated to be generalizable across different backdoor attacks, models, datasets, tasks, and (digital/physical) domains.
Abstract:Despite its prevalent use in image-text matching tasks in a zero-shot manner, CLIP has been shown to be highly vulnerable to adversarial perturbations added onto images. Recent studies propose to finetune the vision encoder of CLIP with adversarial samples generated on the fly, and show improved robustness against adversarial attacks on a spectrum of downstream datasets, a property termed as zero-shot robustness. In this paper, we show that malicious perturbations that seek to maximise the classification loss lead to `falsely stable' images, and propose to leverage the pre-trained vision encoder of CLIP to counterattack such adversarial images during inference to achieve robustness. Our paradigm is simple and training-free, providing the first method to defend CLIP from adversarial attacks at test time, which is orthogonal to existing methods aiming to boost zero-shot adversarial robustness of CLIP. We conduct experiments across 16 classification datasets, and demonstrate stable and consistent gains compared to test-time defence methods adapted from existing adversarial robustness studies that do not rely on external networks, without noticeably impairing performance on clean images. We also show that our paradigm can be employed on CLIP models that have been adversarially finetuned to further enhance their robustness at test time. Our code is available \href{https://github.com/Sxing2/CLIP-Test-time-Counterattacks}{here}.
Abstract:Transferable adversarial examples are known to cause threats in practical, black-box attack scenarios. A notable approach to improving transferability is using integrated gradients (IG), originally developed for model interpretability. In this paper, we find that existing IG-based attacks have limited transferability due to their naive adoption of IG in model interpretability. To address this limitation, we focus on the IG integration path and refine it in three aspects: multiplicity, monotonicity, and diversity, supported by theoretical analyses. We propose the Multiple Monotonic Diversified Integrated Gradients (MuMoDIG) attack, which can generate highly transferable adversarial examples on different CNN and ViT models and defenses. Experiments validate that MuMoDIG outperforms the latest IG-based attack by up to 37.3\% and other state-of-the-art attacks by 8.4\%. In general, our study reveals that migrating established techniques to improve transferability may require non-trivial efforts. Code is available at \url{https://github.com/RYC-98/MuMoDIG}.
Abstract:Recent studies have shown that large vision-language models (LVLMs) often suffer from the issue of object hallucinations (OH). To mitigate this issue, we introduce an efficient method that edits the model weights based on an unsafe subspace, which we call HalluSpace in this paper. With truthful and hallucinated text prompts accompanying the visual content as inputs, the HalluSpace can be identified by extracting the hallucinated embedding features and removing the truthful representations in LVLMs. By orthogonalizing the model weights, input features will be projected into the Null space of the HalluSpace to reduce OH, based on which we name our method Nullu. We reveal that HalluSpaces generally contain statistical bias and unimodal priors of the large language models (LLMs) applied to build LVLMs, which have been shown as essential causes of OH in previous studies. Therefore, null space projection suppresses the LLMs' priors to filter out the hallucinated features, resulting in contextually accurate outputs. Experiments show that our method can effectively mitigate OH across different LVLM families without extra inference costs and also show strong performance in general LVLM benchmarks. Code is released at \url{https://github.com/Ziwei-Zheng/Nullu}.
Abstract:Targeted poisoning attacks aim to compromise the model's prediction on specific target samples. In a common clean-label setting, they are achieved by slightly perturbing a subset of training samples given access to those specific targets. Despite continuous efforts, it remains unexplored whether such attacks can generalize to unknown variations of those targets. In this paper, we take the first step to systematically study this generalization problem. Observing that the widely adopted, cosine similarity-based attack exhibits limited generalizability, we propose a well-generalizable attack that leverages both the direction and magnitude of model gradients. In particular, we explore diverse target variations, such as an object with varied viewpoints and an animal species with distinct appearances. Extensive experiments across various generalization scenarios demonstrate that our method consistently achieves the best attack effectiveness. For example, our method outperforms the cosine similarity-based attack by 20.95% in attack success rate with similar overall accuracy, averaged over four models on two image benchmark datasets. The code is available at https://github.com/jiaangk/generalizable_tcpa
Abstract:Machine learning (ML) has demonstrated significant advancements in Android malware detection (AMD); however, the resilience of ML against realistic evasion attacks remains a major obstacle for AMD. One of the primary factors contributing to this challenge is the scarcity of reliable generalizations. Malware classifiers with limited generalizability tend to overfit spurious correlations derived from biased features. Consequently, adversarial examples (AEs), generated by evasion attacks, can modify these features to evade detection. In this study, we propose a domain adaptation technique to improve the generalizability of AMD by aligning the distribution of malware samples and AEs. Specifically, we utilize meaningful feature dependencies, reflecting domain constraints in the feature space, to establish a robust feature space. Training on the proposed robust feature space enables malware classifiers to learn from predefined patterns associated with app functionality rather than from individual features. This approach helps mitigate spurious correlations inherent in the initial feature space. Our experiments conducted on DREBIN, a renowned Android malware detector, demonstrate that our approach surpasses the state-of-the-art defense, Sec-SVM, when facing realistic evasion attacks. In particular, our defense can improve adversarial robustness by up to 55% against realistic evasion attacks compared to Sec-SVM.
Abstract:Despite prior safety alignment efforts, mainstream LLMs can still generate harmful and unethical content when subjected to jailbreaking attacks. Existing jailbreaking methods fall into two main categories: template-based and optimization-based methods. The former requires significant manual effort and domain knowledge, while the latter, exemplified by Greedy Coordinate Gradient (GCG), which seeks to maximize the likelihood of harmful LLM outputs through token-level optimization, also encounters several limitations: requiring white-box access, necessitating pre-constructed affirmative phrase, and suffering from low efficiency. In this paper, we present ECLIPSE, a novel and efficient black-box jailbreaking method utilizing optimizable suffixes. Drawing inspiration from LLMs' powerful generation and optimization capabilities, we employ task prompts to translate jailbreaking goals into natural language instructions. This guides the LLM to generate adversarial suffixes for malicious queries. In particular, a harmfulness scorer provides continuous feedback, enabling LLM self-reflection and iterative optimization to autonomously and efficiently produce effective suffixes. Experimental results demonstrate that ECLIPSE achieves an average attack success rate (ASR) of 0.92 across three open-source LLMs and GPT-3.5-Turbo, significantly surpassing GCG in 2.4 times. Moreover, ECLIPSE is on par with template-based methods in ASR while offering superior attack efficiency, reducing the average attack overhead by 83%.
Abstract:Deep generative models have demonstrated impressive performance in various computer vision applications, including image synthesis, video generation, and medical analysis. Despite their significant advancements, these models may be used for malicious purposes, such as misinformation, deception, and copyright violation. In this paper, we provide a systematic and timely review of research efforts on defenses against AI-generated visual media, covering detection, disruption, and authentication. We review existing methods and summarize the mainstream defense-related tasks within a unified passive and proactive framework. Moreover, we survey the derivative tasks concerning the trustworthiness of defenses, such as their robustness and fairness. For each task, we formulate its general pipeline and propose a taxonomy based on methodological strategies that are uniformly applicable to the primary subtasks. Additionally, we summarize the commonly used evaluation datasets, criteria, and metrics. Finally, by analyzing the reviewed studies, we provide insights into current research challenges and suggest possible directions for future research.
Abstract:Recent research in adversarial machine learning has focused on visual perception in Autonomous Driving (AD) and has shown that printed adversarial patches can attack object detectors. However, it is important to note that AD visual perception encompasses more than just object detection; it also includes Multiple Object Tracking (MOT). MOT enhances the robustness by compensating for object detection errors and requiring consistent object detection results across multiple frames before influencing tracking results and driving decisions. Thus, MOT makes attacks on object detection alone less effective. To attack such robust AD visual perception, a digital hijacking attack has been proposed to cause dangerous driving scenarios. However, this attack has limited effectiveness. In this paper, we introduce a novel physical-world adversarial patch attack, ControlLoc, designed to exploit hijacking vulnerabilities in entire AD visual perception. ControlLoc utilizes a two-stage process: initially identifying the optimal location for the adversarial patch, and subsequently generating the patch that can modify the perceived location and shape of objects with the optimal location. Extensive evaluations demonstrate the superior performance of ControlLoc, achieving an impressive average attack success rate of around 98.1% across various AD visual perceptions and datasets, which is four times greater effectiveness than the existing hijacking attack. The effectiveness of ControlLoc is further validated in physical-world conditions, including real vehicle tests under different conditions such as outdoor light conditions with an average attack success rate of 77.5%. AD system-level impact assessments are also included, such as vehicle collision, using industry-grade AD systems and production-grade AD simulators with an average vehicle collision rate and unnecessary emergency stop rate of 81.3%.