Deep neural networks (DNNs) are vulnerable to adversarial noises. Adversarial training is a general and effective strategy to improve DNN robustness (i.e., accuracy on noisy data) against adversarial noises. However, DNN models trained by the current existing adversarial training methods may have much lower standard accuracy (i.e., accuracy on clean data), compared to the same models trained by the standard method on clean data, and this phenomenon is known as the trade-off between accuracy and robustness and is considered unavoidable. This issue prevents adversarial training from being used in many application domains, such as medical image analysis, as practitioners do not want to sacrifice standard accuracy too much in exchange for adversarial robustness. Our objective is to lift (i.e., alleviate or even avoid) this trade-off between standard accuracy and adversarial robustness for medical image classification and segmentation. We propose a novel adversarial training method, named Increasing-Margin Adversarial (IMA) Training, which is supported by an equilibrium state analysis about the optimality of adversarial training samples. Our method aims to preserve accuracy while improving robustness by generating optimal adversarial training samples. We evaluate our method and the other eight representative methods on six publicly available image datasets corrupted by noises generated by AutoAttack and white-noise attack. Our method achieves the highest adversarial robustness for image classification and segmentation with the smallest reduction in accuracy on clean data. For one of the applications, our method improves both accuracy and robustness. Our study has demonstrated that our method can lift the trade-off between standard accuracy and adversarial robustness for the image classification and segmentation applications.
The adversarial vulnerability of deep neural networks (DNNs) has been actively investigated in the past several years. This paper investigates the scale-variant property of cross-entropy loss, which is the most commonly used loss function in classification tasks, and its impact on the effective margin and adversarial robustness of deep neural networks. Since the loss function is not invariant to logit scaling, increasing the effective weight norm will make the loss approach zero and its gradient vanish while the effective margin is not adequately maximized. On typical DNNs, we demonstrate that…
Borealis AI；University of Tuebingen（图宾根大学）；Max Planck Institute for Intelligent Systems（马克斯·普朗克智能系统研究所）
We study adversarial robustness of neural networks from a margin maximization perspective, where margins are defined as the distances from inputs to a classifier's decision boundary. Our study shows that maximizing margins can be achieved by minimizing the adversarial loss on the decision boundary at the "shortest successful perturbation", demonstrating a close connection between adversarial losses and the margins. We propose Max-Margin Adversarial (MMA) training to directly maximize the margins to achieve adversarial robustness. Instead of adversarial training with a fixed $ε$, MMA offers an …
Zhejiang University of Science and Technology（浙江科技学院）；Institute of Software, Chinese Academy of Sciences（中国科学院软件研究所）；Zhejiang University（浙江大学）；INSA De Rennes（雷恩国立应用科学学院）；CNRS（法国国家科学研究中心）
Adversarial training, as one of the most effective defense methods against adversarial attacks, tends to learn an inclusive decision boundary to increase the robustness of deep learning models. However, due to the large and unnecessary increase in the margin along adversarial directions, adversarial training causes heavy cross-over between natural examples and adversarial examples, which is not conducive to balancing the trade-off between robustness and natural accuracy. In this paper, we propose a novel adversarial training scheme to achieve a better trade-off between robustness and natural a…
Adversarial training is a technique for training robust machine learning models. To encourage robustness, it iteratively computes adversarial examples for the model, and then re-trains on these examples via some update rule. This work analyzes the performance of adversarial training on linearly separable data, and provides bounds on the number of iterations required for large margin. We show that when the update rule is given by an arbitrary empirical risk minimizer, adversarial training may require exponentially many iterations to obtain large margin. However, if gradient or stochastic gradie…
Advances in the development of adversarial attacks have been fundamental to the progress of adversarial defense research. Efficient and effective attacks are crucial for reliable evaluation of defenses, and also for developing robust models. Adversarial attacks are often generated by maximizing standard losses such as the cross-entropy loss or maximum-margin loss within a constraint set using Projected Gradient Descent (PGD). In this work, we introduce a relaxation term to the standard loss, that finds more suitable gradient-directions, increases attack efficacy and leads to more efficient adv…
Xiamen University（厦门大学）；University of Trento（特伦托大学）
Previous works have extensively studied the transferability of adversarial samples in untargeted black-box scenarios. However, it still remains challenging to craft targeted adversarial examples with higher transferability than non-targeted ones. Recent studies reveal that the traditional Cross-Entropy (CE) loss function is insufficient to learn transferable targeted adversarial examples due to the issue of vanishing gradient. In this work, we provide a comprehensive investigation of the CE loss function and find that the logit margin between the targeted and untargeted classes will quickly ob…
MIT-IBM Watson AI Lab（麻省理工学院-IBM沃森人工智能实验室）；IBM Research（IBM研究院）；University of Illinois, Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）
There are two major paradigms of white-box adversarial attacks that attempt to impose input perturbations. The first paradigm, called the fix-perturbation attack, crafts adversarial samples within a given perturbation level. The second paradigm, called the zero-confidence attack, finds the smallest perturbation needed to cause mis-classification, also known as the margin of an input feature. While the former paradigm is well-resolved, the latter is not. Existing zero-confidence attacks either introduce significant ap-proximation errors, or are too time-consuming. We therefore propose MARGINATT…
Facebook AI Research（Facebook人工智能研究院）；University of Maryland（马里兰大学）；Georgia Institute of Technology（佐治亚理工学院）
Adversarial training is by far the most successful strategy for improving robustness of neural networks to adversarial attacks. Despite its success as a defense mechanism, adversarial training fails to generalize well to unperturbed test set. We hypothesize that this poor generalization is a consequence of adversarial training with uniform perturbation radius around every training sample. Samples close to decision boundary can be morphed into a different class under a small perturbation budget, and enforcing large margins around these samples produce poor decision boundaries that generalize po…
The Chinese University of Hong Kong（香港中文大学）；The University of Melbourne（墨尔本大学）；RIKEN-AIP（理化学研究所人工智能研究中心）；Hong Kong Baptist University（香港浸会大学）
The AutoAttack (AA) has been the most reliable method to evaluate adversarial robustness when considerable computational resources are available. However, the high computational cost (e.g., 100 times more than that of the project gradient descent attack) makes AA infeasible for practitioners with limited computational resources, and also hinders applications of AA in the adversarial training (AT). In this paper, we propose a novel method, minimum-margin (MM) attack, to fast and reliably evaluate adversarial robustness. Compared with AA, our method achieves comparable performance but only costs…
Deep models, while being extremely versatile and accurate, are vulnerable to adversarial attacks: slight perturbations that are imperceptible to humans can completely flip the prediction of deep models. Many attack and defense mechanisms have been proposed, although a satisfying solution still largely remains elusive. In this work, we give strong evidence that during training, deep models maximize the minimum margin in order to achieve high accuracy, but at the same time decrease the \emph{average} margin hence hurting robustness. Our empirical results highlight an intrinsic trade-off between …
Hong Kong Baptist University（香港浸会大学）；University of Technology Sydney（悉尼科技大学）；The University of Sydney（悉尼大学）；Nanjing University of Science and Technology（南京理工大学）；RIKEN Center for Advanced Intelligence Project（理化学研究所先进智能项目中心）；The University of Texas at Austin（德克萨斯大学奥斯汀分校）；The University of Tokyo（东京大学）
Reweighting adversarial data during training has been recently shown to improve adversarial robustness, where data closer to the current decision boundaries are regarded as more critical and given larger weights. However, existing methods measuring the closeness are not very reliable: they are discrete and can take only a few values, and they are path-dependent, i.e., they may change given the same start and end points with different attack paths. In this paper, we propose three types of probabilistic margin (PM), which are continuous and path-independent, for measuring the aforementioned clos…
Nanyang Technological University（南洋理工大学）；University of California, Berkeley（加州大学伯克利分校）
Domain adaptation tackles the problem of transferring knowledge from a label-rich source domain to a label-scarce or even unlabeled target domain. Recently domain-adversarial training (DAT) has shown promising capacity to learn a domain-invariant feature space by reversing the gradient propagation of a domain classifier. However, DAT is still vulnerable in several aspects including (1) training instability due to the overwhelming discriminative ability of the domain classifier in adversarial training, (2) restrictive feature-level alignment, and (3) lack of interpretability or systematic expla…
Adversarial robustness is considered as a required property of deep neural networks. In this study, we discover that adversarially trained models might have significantly different characteristics in terms of margin and smoothness, even they show similar robustness. Inspired by the observation, we investigate the effect of different regularizers and discover the negative effect of the smoothness regularizer on maximizing the margin. Based on the analyses, we propose a new method called bridged adversarial training that mitigates the negative effect by bridging the gap between clean and adversa…
Fudan University（复旦大学）；The University of Melbourne（墨尔本大学）
Evaluating the robustness of a defense model is a challenging task in adversarial robustness research. Obfuscated gradients have previously been found to exist in many defense methods and cause a false signal of robustness. In this paper, we identify a more subtle situation called Imbalanced Gradients that can also cause overestimated adversarial robustness. The phenomenon of imbalanced gradients occurs when the gradient of one term of the margin loss dominates and pushes the attack towards to a suboptimal direction. To exploit imbalanced gradients, we formulate a Margin Decomposition (MD) att…
Does simple trump complex? Comparing strategies for adversarial robustness in DNNs
Faculty of Engineering, North-West University, South Africa（南非北开普大学工程学院）；Centre for Artificial Intelligence Research, South Africa（南非人工智能研究中心）；National Institute for Theoretical and Computational Sciences, South Africa（南非理论与计算科学国家研究所）
University of California San Diego（加州大学圣地亚哥分校）；Halicioǧlu Data Science Institute（哈利西奥卢数据科学研究所）
There has been great interest in enhancing the robustness of neural network classifiers to defend against adversarial perturbations through adversarial training, while balancing the trade-off between robust accuracy and standard accuracy. We propose a novel adversarial training framework that learns to reweight the loss associated with individual training samples based on a notion of class-conditioned margin, with the goal of improving robust generalization. We formulate weighted adversarial training as a bilevel optimization problem with the upper-level problem corresponding to learning a rob…
Mila（米拉研究所）；Université de Montréal（蒙特利尔大学）；University of Waterloo（滑铁卢大学）；Carnegie Mellon University（卡内基梅隆大学）；Google Research（谷歌研究院）
In the context of adversarial robustness, a single model does not usually have enough power to defend against all possible adversarial attacks, and as a result, has sub-optimal robustness. Consequently, an emerging line of work has focused on learning an ensemble of neural networks to defend against adversarial attacks. In this work, we take a principled approach towards building robust ensembles. We view this problem from the perspective of margin-boosting and develop an algorithm for learning an ensemble with maximum margin. Through extensive empirical evaluation on benchmark datasets, we sh…
EPFL（洛桑联邦理工学院）；Adobe Research（Adobe研究院）；University of Surrey（萨里大学）
Image attribution -- matching an image back to a trusted source -- is an emerging tool in the fight against online misinformation. Deep visual fingerprinting models have recently been explored for this purpose. However, they are not robust to tiny input perturbations known as adversarial examples. First we illustrate how to generate valid adversarial images that can easily cause incorrect image attribution. Then we describe an approach to prevent imperceptible adversarial attacks on deep visual fingerprinting models, via robust contrastive learning. The proposed training procedure leverages tr…
University of Maryland, College Park（马里兰大学帕克分校）；Fudan University（复旦大学）；New York University（纽约大学）
Adversarial training is the industry standard for producing models that are robust to small adversarial perturbations. However, machine learning practitioners need models that are robust to other kinds of changes that occur naturally, such as changes in the style or illumination of input images. Such changes in input distribution have been effectively modeled as shifts in the mean and variance of deep image features. We adapt adversarial training by directly perturbing feature statistics, rather than image pixels, to produce models that are robust to various unseen distributional shifts. We ex…
Rutgers University（罗格斯大学）；Google Research（谷歌研究院）；University of Delaware（特拉华大学）
Adversarial data augmentation has shown promise for training robust deep neural networks against unforeseen data shifts or corruptions. However, it is difficult to define heuristics to generate effective fictitious target distributions containing "hard" adversarial perturbations that are largely different from the source distribution. In this paper, we propose a novel and effective regularization term for adversarial data augmentation. We theoretically derive it from the information bottleneck principle, which results in a maximum-entropy formulation. Intuitively, this regularization term enco…
Univeristy of Maryland（马里兰大学）；Weizmann Institute of Science（魏茨曼科学研究所）
Shift invariance is a critical property of CNNs that improves performance on classification. However, we show that invariance to circular shifts can also lead to greater sensitivity to adversarial attacks. We first characterize the margin between classes when a shift-invariant linear classifier is used. We show that the margin can only depend on the DC component of the signals. Then, using results about infinitely wide networks, we show that in some simple cases, fully connected and shift-invariant neural networks produce linear decision boundaries. Using this, we prove that shift invariance i…
Dataset bias is a problem in adversarial machine learning, especially in the evaluation of defenses. An adversarial attack or defense algorithm may show better results on the reported dataset than can be replicated on other datasets. Even when two algorithms are compared, their relative performance can vary depending on the dataset. Deep learning offers state-of-the-art solutions for image recognition, but deep models are vulnerable even to small perturbations. Research in this area focuses primarily on adversarial attacks and defense algorithms. In this paper, we report for the first time, a …
Adversarial training has been widely acknowledged as the most effective method to improve the adversarial robustness against adversarial examples for Deep Neural Networks (DNNs). So far, most existing works focus on enhancing the overall model robustness, treating each class equally in both the training and testing phases. Although revealing the disparity in robustness among classes, few works try to make adversarial training fair at the class level without sacrificing overall robustness. In this paper, we are the first to theoretically and empirically investigate the preference of different c…
Deep neural networks are vulnerable to adversarial examples -- minor perturbations added to a model's input which cause the model to output an incorrect prediction. We introduce a new method for improving the efficacy of adversarial attacks in a black-box setting by undertraining the surrogate model which the attacks are generated on. Using two datasets and five model architectures, we show that this method transfers well across architectures and outperforms state-of-the-art methods by a wide margin. We interpret the effectiveness of our approach as a function of reduced surrogate model loss f…
Montreal Institute for Learning Algorithms（蒙特利尔学习算法研究所）；University of Montreal（蒙特利尔大学）；York University（约克大学）
Gatys et al. (2015) showed that optimizing pixels to match features in a convolutional network with respect reference image features is a way to render images of high visual quality. We show that unrolling this gradient-based optimization yields a recurrent computation that creates images by incrementally adding onto a visual "canvas". We propose a recurrent generative model inspired by this view, and show that it can be trained using adversarial training to generate very good image samples. We also propose a way to quantitatively compare adversarial networks by having the generators and discr…
Medical imaging AI systems such as disease classification and segmentation are increasingly inspired and transformed from computer vision based AI systems. Although an array of adversarial training and/or loss function based defense techniques have been developed and proved to be effective in computer vision, defending against adversarial attacks on medical images remains largely an uncharted territory due to the following unique challenges: 1) label scarcity in medical images significantly limits adversarial generalizability of the AI system; 2) vastly similar and dominant fore- and backgroun…
As the name suggests, image spam is spam email that has been embedded in an image. Image spam was developed in an effort to evade text-based filters. Modern deep learning-based classifiers perform well in detecting typical image spam that is seen in the wild. In this chapter, we evaluate numerous adversarial techniques for the purpose of attacking deep learning-based image spam classifiers. Of the techniques tested, we find that universal perturbation performs best. Using universal adversarial perturbations, we propose and analyze a new transformation-based adversarial attack that enables us t…
It is known that Deep Neural networks (DNNs) are vulnerable to adversarial attacks, and the adversarial robustness of DNNs could be improved by adding adversarial noises to training data (e.g., the standard adversarial training (SAT)). However, inappropriate noises added to training data may reduce a model's performance, which is termed the trade-off between accuracy and robustness. This problem has been sufficiently studied for the classification of whole images but has rarely been explored for image analysis tasks in the medical application domain, including image segmentation, landmark dete…
Transferable Adversarial Robustness for Speech Foundation Models via Hierarchical Stabilization
Sharif University of Technology（谢里夫理工大学）；University of Tehran（德黑兰大学）；New Uzbekistan University（新乌兹别克斯坦大学）；Okinawa Institute of Science and Technology（冲绳科学技术大学院大学）
This work systematically investigates the adversarial robustness of deep image denoisers (DIDs), i.e, how well DIDs can recover the ground truth from noisy observations degraded by adversarial perturbations. Firstly, to evaluate DIDs' robustness, we propose a novel adversarial attack, namely Observation-based Zero-mean Attack ({\sc ObsAtk}), to craft adversarial zero-mean perturbations on given noisy images. We find that existing DIDs are vulnerable to the adversarial noise generated by {\sc ObsAtk}. Secondly, to robustify DIDs, we propose an adversarial training strategy, hybrid adversarial t…
Transductive Adversarial Networks (TAN) is a novel domain-adaptation machine learning framework that is designed for learning a conditional probability distribution on unlabelled input data in a target domain, while also only having access to: (1) easily obtained labelled data from a related source domain, which may have a different conditional probability distribution than the target domain, and (2) a marginalised prior distribution on the labels for the target domain. TAN leverages a fully adversarial training procedure and a unique generator/encoder architecture which approximates the trans…
Weizmann Institute of Science（魏茨曼科学研究所）；TTIC（芝加哥丰田技术研究所）；Toyoyta Technological Institute at Chicago（芝加哥丰田技术研究所）
We propose an approach to distinguish between correct and incorrect image classifications. Our approach can detect misclassifications which either occur $\it{unintentionally}$ ("natural errors"), or due to $\it{intentional~adversarial~attacks}$ ("adversarial errors"), both in a single $\it{unified~framework}$. Our approach is based on the observation that correctly classified images tend to exhibit robust and consistent classifications under certain image transformations (e.g., horizontal flip, small image translation, etc.). In contrast, incorrectly classified images (whether due to adversari…
Image normalization is a critical step in medical imaging. This step is often done on a per-dataset basis, preventing current segmentation algorithms from the full potential of exploiting jointly normalized information across multiple datasets. To solve this problem, we propose an adversarial normalization approach for image segmentation which learns common normalizing functions across multiple datasets while retaining image realism. The adversarial training provides an optimal normalizer that improves both the segmentation accuracy and the discrimination of unrealistic normalizing functions. …
Recent studies have shown that adversarial examples hand-crafted on one white-box model can be used to attack other black-box models. Such cross-model transferability makes it feasible to perform black-box attacks, which has raised security concerns for real-world DNNs applications. Nevertheless, existing works mostly focus on investigating the adversarial transferability across different deep models that share the same modality of input data. The cross-modal transferability of adversarial perturbation has never been explored. This paper investigates the transferability of adversarial perturba…
Deep convolutional neural networks can be highly vulnerable to small perturbations of their inputs, potentially a major issue or limitation on system robustness when using deep networks as classifiers. In this paper we propose a low-cost method to explore marginal sample data near trained classifier decision boundaries, thus identifying potential adversarial samples. By finding such adversarial samples it is possible to reduce the search space of adversarial attack algorithms while keeping a reasonable successful perturbation rate. In our developed strategy, the potential adversarial samples r…
Anhui University（安徽大学）；University Of Strathclyde（思克莱德大学）；East China Normal University（华东师范大学）
Adding perturbations to images can mislead classification models to produce incorrect results. Recently, researchers exploited adversarial perturbations to protect image privacy from retrieval by intelligent models. However, adding adversarial perturbations to images destroys the original data, making images useless in digital forensics and other fields. To prevent illegal or unauthorized access to sensitive image data such as human faces without impeding legitimate users, the use of reversible adversarial attack techniques is increasing. The original image can be recovered from its reversible…
Progress in making neural networks more robust against adversarial attacks is mostly marginal, despite the great efforts of the research community. Moreover, the robustness evaluation is often imprecise, making it difficult to identify promising approaches. We analyze the classification decisions of 19 different state-of-the-art neural networks trained to be robust against adversarial attacks. Our findings suggest that current untargeted adversarial attacks induce misclassification towards only a limited amount of different classes. Additionally, we observe that both over- and under-confidence…
Deep learning technologies have become the backbone for the development of computer vision. With further explorations, deep neural networks have been found vulnerable to well-designed adversarial attacks. Most of the vision devices are equipped with image signal processing (ISP) pipeline to implement RAW-to-RGB transformations and embedded into data preprocessing module for efficient image processing. Actually, ISP pipeline can introduce adversarial behaviors to post-capture images while data preprocessing may destroy attack patterns. However, none of the existing adversarial attacks takes int…
University of Chinese Academy of Sciences（中国科学院大学）；Center for Research on Intelligent Perception and Computing, CASIA（中国科学院自动化研究所智能感知与计算研究中心）
Images synthesized by powerful generative adversarial network (GAN) based methods have drawn moral and privacy concerns. Although image forensic models have reached great performance in detecting fake images from real ones, these models can be easily fooled with a simple adversarial attack. But, the noise adding adversarial samples are also arousing suspicion. In this paper, instead of adding adversarial noise, we optimally search adversarial points on face manifold to generate anti-forensic fake face images. We iteratively do a gradient-descent with each small step in the latent space of a ge…
The University of Kansas（堪萨斯大学）；The University of Oklahoma（俄克拉荷马大学）
In the evasion attacks against deep neural networks (DNN), the attacker generates adversarial instances that are visually indistinguishable from benign samples and sends them to the target DNN to trigger misclassifications. In this paper, we propose a novel multi-view adversarial image detector, namely Argos, based on a novel observation. That is, there exist two "souls" in an adversarial instance, i.e., the visually unchanged content, which corresponds to the true label, and the added invisible perturbation, which corresponds to the misclassified label. Such inconsistencies could be further a…
Adversarial training is a popular defense strategy against attack threat models with bounded Lp norms. However, it often degrades the model performance on normal images and the defense does not generalize well to novel attacks. Given the success of deep generative models such as GANs and VAEs in characterizing the underlying manifold of images, we investigate whether or not the aforementioned problems can be remedied by exploiting the underlying manifold information. To this end, we construct an "On-Manifold ImageNet" (OM-ImageNet) dataset by projecting the ImageNet samples onto the manifold l…
Ohio State University（俄亥俄州立大学）；California State Polytechnic University, Pomona（加州州立理工大学波莫纳分校）
Today's state-of-the-art image classifiers fail to correctly classify carefully manipulated adversarial images. In this work, we develop a new, localized adversarial attack that generates adversarial examples by imperceptibly altering the backgrounds of normal images. We first use this attack to highlight the unnecessary sensitivity of neural networks to changes in the background of an image, then use it as part of a new training technique: localized adversarial training. By including locally adversarial images in the training set, we are able to create a classifier that suffers less loss than…
Xi’an Jiaotong University（西安交通大学）；University of British Columbia（不列颠哥伦比亚大学）；Beijing Normal University（北京师范大学）
Deep learning based image classification models are shown vulnerable to adversarial attacks by injecting deliberately crafted noises to clean images. To defend against adversarial attacks in a training-free and attack-agnostic manner, this work proposes a novel and effective reconstruction-based defense framework by delving into deep image prior (DIP). Fundamentally different from existing reconstruction-based defenses, the proposed method analyzes and explicitly incorporates the model decision process into our defense. Given an adversarial image, firstly we map its reconstructed images during…
Deep neural networks are vulnerable to adversarial attacks. White-box adversarial attacks can fool neural networks with small adversarial perturbations, especially for large size images. However, keeping successful adversarial perturbations imperceptible is especially challenging for transfer-based black-box adversarial attacks. Often such adversarial examples can be easily spotted due to their unpleasantly poor visual qualities, which compromises the threat of adversarial attacks in practice. In this study, to improve the image quality of black-box adversarial examples perceptually, we propos…
Adversarial Resilience in Sequential Prediction via Abstention
University of Pennsylvania（宾夕法尼亚大学）；Purdue University（普渡大学）；Technion（以色列理工学院）；Google Research, Israel（谷歌研究院（以色列））；University of California, Berkeley（加利福尼亚大学伯克利分校）
National Tsing Hua University（国立清华大学）；The Chinese University of Hong Kong（香港中文大学）；Columbia University（哥伦比亚大学）；IBM Research（IBM研究院）
Model robustness against adversarial examples of single perturbation type such as the $\ell_{p}$-norm has been widely studied, yet its generalization to more realistic scenarios involving multiple semantic perturbations and their composition remains largely unexplored. In this paper, we first propose a novel method for generating composite adversarial examples. Our method can find the optimal attack composition by utilizing component-wise projected gradient descent and automatic attack-order scheduling. We then propose generalized adversarial training (GAT) to extend model robustness from $\el…
