Vision transformers (ViTs) perform exceptionally well in various computer vision tasks but remain vulnerable to adversarial attacks. Recent studies have shown that the transferability of adversarial examples exists fo...
Vision transformers (ViTs) perform exceptionally well in various computer vision tasks but remain vulnerable to adversarial attacks. Recent studies have shown that the transferability of adversarial examples exists for CNNs, and the same holds true for ViTs. However, existing ViT attacks aggressively regularize the largest token gradients to exact zero within each layer of the surrogate model, overlooking the interactions between layers, which limits their transferability in attacking black-box models. Therefore, in this paper, we focus on boosting the transferability of adversarial attacks on ViTs through adaptive token tuning (ATT). Specifically, we propose three optimization strategies: an adaptive gradient re-scaling strategy to reduce the overall variance of token gradients, a self-paced patch out strategy to enhance the diversity of input tokens, and a hybrid token gradient truncation strategy to weaken the effectiveness of attention mechanism. We demonstrate that scaling correction of gradient changes using gradient variance across different layers can produce highly transferable adversarial examples. In addition, introducing attentional truncation can mitigate the overfitting over complex interactions between tokens in deep ViT layers to further improve the transferability. On the other hand, using feature importance as a guidance to discard a subset of perturbation patches in each iteration, along with combining self-paced learning and progressively more sampled attacks, significantly enhances the transferability over attacks that use all perturbation patches. Extensive experiments conducted on ViTs, undefended CNNs, and defended CNNs validate the superiority of our proposed ATT attack method. On average, our approach improves the attack performance by 10.1% compared to state-of-the-art transfer-based attacks. Notably, we achieve the best attack performance with an average of 58.3% on three defended CNNs. Code is available at https://***/MisterRpeng/
IV (intelligent vehicles) lane departure detection and early warning system can improve driving safety, reduce accidents caused by distracted or tired driving, and it is also an important part of automatic driving tec...
详细信息
Pose prediction can timely determine whether the posture of industrial robots is abnormal, thereby reducing the harm caused by abnormal actions of industrial robots. The pose prediction task requires modeling the comp...
详细信息
Fraud detection datasets typically contain a large number of normal transaction samples and very few fraudulent samples, resulting in a severely imbalanced dataset. This imbalance affects the model’s classification a...
详细信息
Hierarchical Text Classification (HTC) is a challenging task which aims to extract the labels in a tree structure corresponding to a given text. Discriminative methods usually incorporate the hierarchical structure in...
详细信息
This paper proposes a comprehensive approach for enhancing security in medical record systems by incorporating One-Time Password (OTP) verification and Zero Trust Security principles within a Role-Based Access Control...
详细信息
Network analysis is necessary to understand the structure and dynamics of complex systems. In this study, we use the Louvain algorithm for community detection in networks and examine the effect of seed size on central...
详细信息
Face Presentation Attack Detection(fPAD)plays a vital role in securing face recognition systems against various presentation *** supervised learning-based methods demonstrate effectiveness,they are prone to overfittin...
详细信息
Face Presentation Attack Detection(fPAD)plays a vital role in securing face recognition systems against various presentation *** supervised learning-based methods demonstrate effectiveness,they are prone to overfitting to known attack types and struggle to generalize to novel attack *** studies have explored formulating fPAD as an anomaly detection problem or one-class classification task,enabling the training of generalized models for unknown attack ***,conventional anomaly detection approaches encounter difficulties in precisely delineating the boundary between bonafide samples and unknown *** address this challenge,we propose a novel framework focusing on unknown attack detection using exclusively bonafide facial data during *** core innovation lies in our pseudo-negative sample synthesis(PNSS)strategy,which facilitates learning of compact decision boundaries between bonafide faces and potential attack ***,PNSS generates synthetic negative samples within low-likelihood regions of the bonafide feature space to represent diverse unknown attack *** overcome the inherent imbalance between positive and synthetic negative samples during iterative training,we implement a dual-loss mechanism combining focal loss for classification optimization with pairwise confusion loss as a *** architecture effectively mitigates model bias towards bonafide samples while maintaining discriminative *** evaluations across three benchmark datasets validate the framework’s superior ***,our PNSS achieves 8%–18% average classification error rate(ACER)reduction compared with state-of-the-art one-class fPAD methods in cross-dataset evaluations on Idiap Replay-Attack and MSU-MFSD datasets.
Essay are considered as the most prominent way in assessing the student performance. Manual grading of the essay are considered to be a hectic tasks for both students and instructors due to various reasons such as tim...
详细信息
Traditional Indian Medicine faces the risk of extinction due to insufficient transfer of knowledge. This research highlights the critical need for preservation through modern technology and techniques. It addresses th...
详细信息
暂无评论