On Robustness and Transferability of Convolutional Neural Networks
Google Research（谷歌研究院）
Josip Djolonga, Jessica Yung, Michael Tschannen, Rob Romijnders, Lucas Beyer, Alexander Kolesnikov, Joan Puigcerver, Matthias Minderer, Alexander D'Amour, Dan M… 展开作者
Josip Djolonga, Jessica Yung, Michael Tschannen, Rob Romijnders, Lucas Beyer, Alexander Kolesnikov, Joan Puigcerver, Matthias Minderer, Alexander D'Amour, Dan Moldovan, Sylvain Gelly, Neil Houlsby, Xiaohua Zhai, Mario Lucic
Modern deep convolutional networks (CNNs) are often criticized for not generalizing under distributional shifts. However, several recent breakthroughs in transfer learning suggest that these networks can cope with severe distribution shifts and successfully adapt to new tasks from a few training examples. In this work we study the interplay between out-of-distribution and transfer performance of modern image classification CNNs for the first time and investigate the impact of the pre-training data size, the model scale, and the data preprocessing pipeline. We find that increasing both the training set and model sizes significantly improve the distributional shift robustness. Furthermore, we show that, perhaps surprisingly, simple changes in the preprocessing such as modifying the image resolution can significantly mitigate robustness issues in some cases. Finally, we outline the shortcomings of existing robustness evaluation datasets and introduce a synthetic dataset SI-Score we use for a systematic analysis across factors of variation common in visual data such as object size and position.
Raw deep neural network (DNN) performance is not enough; in real-world settings, computational load, training efficiency and adversarial security are just as or even more important. We propose to simultaneously tackle Performance, Efficiency, and Robustness, using our proposed algorithm Q-TART, Quickly Train for Adversarial Robustness and in-Transferability. Q-TART follows the intuition that samples highly susceptible to noise strongly affect the decision boundaries learned by DNNs, which in turn degrades their performance and adversarial susceptibility. By identifying and removing such sample…
This paper proposes Characteristic Examples for effectively fingerprinting deep neural networks, featuring high-robustness to the base model against model pruning as well as low-transferability to unassociated models. This is the first work taking both robustness and transferability into consideration for generating realistic fingerprints, whereas current methods lack practical assumptions and may incur large false positive rates. To achieve better trade-off between robustness and transferability, we propose three kinds of characteristic examples: vanilla C-examples, RC-examples, and LTRC-exam…
Monash University（莫纳什大学）；Defence Science and Technology Group（国防科技集团）
Ensemble-based adversarial training is a principled approach to achieve robustness against adversarial attacks. An important technique of this approach is to control the transferability of adversarial examples among ensemble members. We propose in this work a simple yet effective strategy to collaborate among committee models of an ensemble model. This is achieved via the secure and insecure sets defined for each model member on a given sample, hence help us to quantify and regularize the transferability. Consequently, our proposed framework provides the flexibility to reduce the adversarial t…
City University of Hong Kong（香港城市大学）；Dalian University of Technology（大连理工大学）；Tsinghua University（清华大学）
Recent years have seen the ever-increasing importance of pre-trained models and their downstream training in deep learning research and applications. At the same time, the defense for adversarial examples has been mainly investigated in the context of training from random initialization on simple classification tasks. To better exploit the potential of pre-trained models in adversarial robustness, this paper focuses on the fine-tuning of an adversarially pre-trained model in various classification tasks. Existing research has shown that since the robust pre-trained model has already learned a …
University of Illinois at Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）
Machine learning (ML) robustness and domain generalization are fundamentally correlated: they essentially concern data distribution shifts under adversarial and natural settings, respectively. On one hand, recent studies show that more robust (adversarially trained) models are more generalizable. On the other hand, there is a lack of theoretical understanding of their fundamental connections. In this paper, we explore the relationship between regularization and domain transferability considering different factors such as norm regularization and data augmentations (DA). We propose a general the…
Why Does Little Robustness Help? A Further Step Towards Understanding Adversarial Transferability
School of Cyber Science and Engineering, Huazhong University of Science and Technology（电子科学与工程学院，华中科技大学）；School of Software Engineering, Huazhong University of Science and Technology（软件工程学院，华中科技大学）；School of Information and Communication Technology, Griffith University（信息与通信技术学院，格里菲斯大学）；School of Computer Science and Technology, Huazhong University of Science and Technology（计算机科学与技术学院，华中科技大学）
Korea Advanced Institute of Science Technology (KAIST)（韩国科学技术院）
Batch normalization (BN) has been widely used in modern deep neural networks (DNNs) due to improved convergence. BN is observed to increase the model accuracy while at the cost of adversarial robustness. There is an increasing interest in the ML community to understand the impact of BN on DNNs, especially related to the model robustness. This work attempts to understand the impact of BN on DNNs from a non-robust feature perspective. Straightforwardly, the improved accuracy can be attributed to the better utilization of useful features. It remains unclear whether BN mainly favors learning robus…
Physically-Induced Atmospheric Adversarial Perturbations: Enhancing Transferability and Robustness in Remote Sensing Image Classification
School of Computer and Information Engineering, Xiamen University of Technology（厦门理工学院计算机与信息工程学院）；Faculty of Data Science, City University of Macau（澳门城市大学数据科学学院）；Dundee International Institute, Central South University（中南大学 Dundee 国际学院）；Department of Electrical and Computer Engineering, University of Hong Kong（香港大学电气与计算机工程系）；Department of Engineering, University of Cambridge（剑桥大学工程系）；School of Computer Science, Engineering Research Center of Machine Learning and Industry Intelligence, Sichuan University（四川大学计算机科学学院、机器学习与工业智能工程研究中心）；Hong Kong JC STEM Lab of Smart City and Department of Computer Science, City University of Hong Kong（香港城市大学智慧城市JC STEM实验室与计算机科学系）；Department of Computer and Information Science, Faculty of Science and Technology, University of Macau（澳门大学科学与技术学院计算机与信息科学系）
Deep Learning models hold state-of-the-art performance in many fields, but their vulnerability to adversarial examples poses threat to their ubiquitous deployment in practical settings. Additionally, adversarial inputs generated on one classifier have been shown to transfer to other classifiers trained on similar data, which makes the attacks possible even if model parameters are not revealed to the adversary. This property of transferability has not yet been systematically studied, leading to a gap in our understanding of robustness of neural networks to adversarial inputs. In this work, we s…
University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）；Tencent Inc.（腾讯公司）；University of Melbourne（墨尔本大学）；Huazhong University of Science and Technology（华中科技大学）；ETH Zurich（苏黎世联邦理工学院）
Adversarial Transferability is an intriguing property - adversarial perturbation crafted against one model is also effective against another model, while these models are from different model families or training processes. To better protect ML systems against adversarial attacks, several questions are raised: what are the sufficient conditions for adversarial transferability and how to bound it? Is there a way to reduce the adversarial transferability in order to improve the robustness of an ensemble ML model? To answer these questions, in this work we first theoretically analyze and outline …
Virtual high throughput screening (VHTS) and machine learning (ML) have greatly accelerated the design of single-site transition-metal catalysts. VHTS of catalysts, however, is often accompanied with high calculation failure rate and wasted computational resources due to the difficulty of simultaneously converging all mechanistically relevant reactive intermediates to expected geometries and electronic states. We demonstrate a dynamic classifier approach, i.e., a convolutional neural network that monitors geometry optimization on the fly, and exploit its good performance and transferability fo…
Neural network compression methods like pruning and quantization are very effective at efficiently deploying Deep Neural Networks (DNNs) on edge devices. However, DNNs remain vulnerable to adversarial examples-inconspicuous inputs that are specifically designed to fool these models. In particular, Universal Adversarial Perturbations (UAPs), are a powerful class of adversarial attacks which create adversarial perturbations that can generalize across a large set of inputs. In this work, we analyze the effect of various compression techniques to UAP attacks, including different forms of pruning a…
Adversarial examples, which are slightly perturbed inputs generated with the aim of fooling a neural network, are known to transfer between models; adversaries which are effective on one model will often fool another. This concept of transferability poses grave security concerns as it leads to the possibility of attacking models in a black box setting, during which the internal parameters of the target model are unknown. In this paper, we seek to analyze and minimize the transferability of adversaries between models within an ensemble. To this end, we introduce a gradient based measure of how …
Australian National University（澳大利亚国立大学）；Stony Brook University（石溪大学）；Linköping University（林雪平大学）；Mohamed bin Zayed University of AI（穆罕默德·本·扎耶德人工智能大学）；Qualcomm USA（美国高通公司）
Vision transformers (ViTs) process input images as sequences of patches via self-attention; a radically different architecture than convolutional neural networks (CNNs). This makes it interesting to study the adversarial feature space of ViT models and their transferability. In particular, we observe that adversarial patterns found via conventional adversarial attacks show very \emph{low} black-box transferability even for large ViT models. We show that this phenomenon is only due to the sub-optimal attack procedures that do not leverage the true representation potential of ViTs. A deep ViT is…
Though deep neural networks perform challenging tasks excellently, they are susceptible to adversarial examples, which mislead classifiers by applying human-imperceptible perturbations on clean inputs. Under the query-free black-box scenario, adversarial examples are hard to transfer to unknown models, and several methods have been proposed with the low transferability. To settle such issue, we design a max-min framework inspired by input transformations, which are benificial to both the adversarial attack and defense. Explicitly, we decrease loss values with inputs' affline transformations as…
Massachusetts Institute of Technology（麻省理工学院）；Michigan State University（密歇根州立大学）；MIT-IBM Watson AI Lab, IBM Research（MIT-IBM沃森人工智能实验室，IBM研究院）
Contrastive learning (CL) can learn generalizable feature representations and achieve the state-of-the-art performance of downstream tasks by finetuning a linear classifier on top of it. However, as adversarial robustness becomes vital in image classification, it remains unclear whether or not CL is able to preserve robustness to downstream tasks. The main challenge is that in the self-supervised pretraining + supervised finetuning paradigm, adversarial robustness is easily forgotten due to a learning task mismatch from pretraining to finetuning. We call such a challenge 'cross-task robustness…
Ghent University（根特大学）；VIB Inflammation Research Center（VIB炎症研究中心）；imec（比利时微电子研究中心）
Regional adversarial attacks often rely on complicated methods for generating adversarial perturbations, making it hard to compare their efficacy against well-known attacks. In this study, we show that effective regional perturbations can be generated without resorting to complex methods. We develop a very simple regional adversarial perturbation attack method using cross-entropy sign, one of the most commonly used losses in adversarial machine learning. Our experiments on ImageNet with multiple models reveal that, on average, $76\%$ of the generated adversarial examples maintain model-to-mode…
Sabanci University（萨班哲大学）；University of Padua（帕多瓦大学）；Moulay Ismail University of Meknes（梅克内斯穆莱·伊斯梅尔大学）；Florida International University（佛罗里达国际大学）
Convolutional Neural Networks (CNNs) models are one of the most frequently used deep learning networks, and extensively used in both academia and industry. Recent studies demonstrated that adversarial attacks against such models can maintain their effectiveness even when used on models other than the one targeted by the attacker. This major property is known as transferability, and makes CNNs ill-suited for security applications. In this paper, we provide the first comprehensive study which assesses the robustness of CNN-based models for computer networks against adversarial transferability. F…
In recent years, convolutional neural networks (CNNs) have been widely used by researchers to perform forensic tasks such as image tampering detection. At the same time, adversarial attacks have been developed that are capable of fooling CNN-based classifiers. Understanding the transferability of adversarial attacks, i.e. an attacks ability to attack a different CNN than the one it was trained against, has important implications for designing CNNs that are resistant to attacks. While attacks on object recognition CNNs are believed to be transferrable, recent work by Barni et al. has shown that…
In the last decade, the use of Machine Learning techniques in anomaly-based intrusion detection systems has seen much success. However, recent studies have shown that Machine learning in general and deep learning specifically are vulnerable to adversarial attacks where the attacker attempts to fool models by supplying deceptive input. Research in computer vision, where this vulnerability was first discovered, has shown that adversarial images designed to fool a specific model can deceive other machine learning models. In this paper, we investigate the transferability of adversarial network tra…
RealAI（瑞莱智慧）；Peng Cheng Laboratory（鹏城实验室）；Pazhou Laboratory（琶洲实验室）；Tsinghua University（清华大学）；Sea AI Lab（Sea人工智能实验室）
Transfer-based adversarial attacks can evaluate model robustness in the black-box setting. Several methods have demonstrated impressive untargeted transferability, however, it is still challenging to efficiently produce targeted transferability. To this end, we develop a simple yet effective framework to craft targeted transfer-based adversarial examples, applying a hierarchical generative network. In particular, we contribute to amortized designs that well adapt to multi-class targeted attacks. Extensive experiments on ImageNet show that our method improves the success rates of targeted black…
Rice University（莱斯大学）；MIT-IBM Watson AI Lab（麻省理工学院-IBM沃森人工智能实验室）；Georgia Institute of Technology（佐治亚理工学院）；University of Virginia（弗吉尼亚大学）
Generalized Zero-Shot Learning (GZSL) aims to train a classifier that can generalize to unseen classes, using a set of attributes as auxiliary information, and the visual features extracted from a pre-trained convolutional neural network. While recent GZSL methods have explored various techniques to leverage the capacity of these features, there has been an extensive growth of representation learning techniques that remain under-explored. In this work, we investigate the utility of different GZSL methods when using different feature extractors, and examine how these models' pre-training object…
New York University（纽约大学）；Facebook AI Research（Facebook人工智能研究）；CIFAR（加拿大高级研究所（CIFAR））
We propose a retrieval-augmented convolutional network and propose to train it with local mixup, a novel variant of the recently proposed mixup algorithm. The proposed hybrid architecture combining a convolutional network and an off-the-shelf retrieval engine was designed to mitigate the adverse effect of off-manifold adversarial examples, while the proposed local mixup addresses on-manifold ones by explicitly encouraging the classifier to locally behave linearly on the data manifold. Our evaluation of the proposed approach against five readily-available adversarial attacks on three datasets--…
Recently, the generalization behavior of Convolutional Neural Networks (CNN) is gradually transparent through explanation techniques with the frequency components decomposition. However, the importance of the phase spectrum of the image for a robust vision system is still ignored. In this paper, we notice that the CNN tends to converge at the local optimum which is closely related to the high-frequency components of the training images, while the amplitude spectrum is easily disturbed such as noises or common corruptions. In contrast, more empirical studies found that humans rely on more phase…
EAGT: Echocardiography Augmentation for Generalisability and Transferability
THRIVE Centre, University of West London（西伦敦大学THRIVE中心）；University of West London（西伦敦大学）；School of Computing and Engineering, University of West London（西伦敦大学计算机与工程学院）
Johns Hopkins University（约翰斯·霍普金斯大学）；University of Oxford（牛津大学）；Baidu Research（百度研究院）；Wormpex AI Research（Wormpex人工智能研究院）
Though CNNs have achieved the state-of-the-art performance on various vision tasks, they are vulnerable to adversarial examples --- crafted by adding human-imperceptible perturbations to clean images. However, most of the existing adversarial attacks only achieve relatively low success rates under the challenging black-box setting, where the attackers have no knowledge of the model structure and parameters. To this end, we propose to improve the transferability of adversarial examples by creating diverse input patterns. Instead of only using the original images to generate adversarial examples…
This paper focuses on high-transferable adversarial attacks on detectors, which are hard to attack in a black-box manner, because of their multiple-output characteristics and the diversity across architectures. To pursue a high attack transferability, one plausible way is to find a common property across detectors, which facilitates the discovery of common weaknesses. We are the first to suggest that the relevance map from interpreters for detectors is such a property. Based on it, we design a Relevance Attack on Detectors (RAD), which achieves a state-of-the-art transferability, exceeding exi…
McGill University（麦吉尔大学）；École de Technologie Superiéure, Université du Québec（魁北克大学高等技术学院）；Pontifical Catholic University of Paraná（巴拉那天主教大学）
This paper shows the susceptibility of spectrogram-based audio classifiers to adversarial attacks and the transferability of such attacks to audio waveforms. Some commonly used adversarial attacks to images have been applied to Mel-frequency and short-time Fourier transform spectrograms, and such perturbed spectrograms are able to fool a 2D convolutional neural network (CNN). Such attacks produce perturbed spectrograms that are visually imperceptible by humans. Furthermore, the audio waveforms reconstructed from the perturbed spectrograms are also able to fool a 1D CNN trained on the original …
Machine learning technologies using deep neural networks (DNNs), especially convolutional neural networks (CNNs), have made automated, accurate, and fast medical image analysis a reality for many applications, and some DNN-based medical image analysis systems have even been FDA-cleared. Despite the progress, challenges remain to build DNNs as reliable as human expert doctors. It is known that DNN classifiers may not be robust to noises: by adding a small amount of noise to an input image, a DNN classifier may make a wrong classification of the noisy image (i.e., in-distribution adversarial sam…
Facebook AI Research（脸书人工智能研究院）；Univ. Grenoble Alpes（格勒诺布尔阿尔卑斯大学）；Inria（法国国家信息与自动化研究所）；CNRS（法国国家科学研究中心）；Grenoble INP（格勒诺布尔国立理工学院）
Convolutional neural networks trained without supervision come close to matching performance with supervised pre-training, but sometimes at the cost of an even higher number of parameters. Extracting subnetworks from these large unsupervised convnets with preserved performance is of particular interest to make them less computationally intensive. Typical pruning methods operate during training on a task while trying to maintain the performance of the pruned network on the same task. However, in self-supervised feature learning, the training objective is agnostic on the representation transfera…
Offenburg University（奥芬堡大学）；Fraunhofer ITWM（弗劳恩霍夫工业数学研究所）；Fraunhofer Research Center ML（弗劳恩霍夫机器学习研究中心）
Deep learning models are intrinsically sensitive to distribution shifts in the input data. In particular, small, barely perceivable perturbations to the input data can force models to make wrong predictions with high confidence. An common defense mechanism is regularization through adversarial training which injects worst-case perturbations back into training to strengthen the decision boundaries, and to reduce overfitting. In this context, we perform an investigation of 3x3 convolution filters that form in adversarially-trained models. Filters are extracted from 71 public models of the linf-R…
Breaking Adversarial Transferability in Fine-Tuned Speech Recognition
Idiap Research Institute（Idiap研究所）；École Polytechnique Fédérale de Lausanne (EPFL)（洛桑联邦理工学院）；Sharif University of Technology（谢里夫理工大学）；University of Isfahan（伊斯法罕大学）
Bahçeşehir University（巴赫切谢希尔大学）；University of Padua（帕多瓦大学）；Florida International University（佛罗里达国际大学）
In the past decades, the rise of artificial intelligence has given us the capabilities to solve the most challenging problems in our day-to-day lives, such as cancer prediction and autonomous navigation. However, these applications might not be reliable if not secured against adversarial attacks. In addition, recent works demonstrated that some adversarial examples are transferable across different models. Therefore, it is crucial to avoid such transferability via robust models that resist adversarial manipulations. In this paper, we propose a feature randomization-based approach that resists …
University Of Texas at Arlington（德克萨斯大学阿灵顿分校）；Kuaishou Techlology（快手科技）
Unsupervised domain adaptation (UDA) aims to transfer the knowledge learnt from a labeled source domain to an unlabeled target domain. Previous work is mainly built upon convolutional neural networks (CNNs) to learn domain-invariant representations. With the recent exponential increase in applying Vision Transformer (ViT) to vision tasks, the capability of ViT in adapting cross-domain knowledge, however, remains unexplored in the literature. To fill this gap, this paper first comprehensively investigates the transferability of ViT on a variety of domain adaptation tasks. Surprisingly, ViT demo…
Robustness of neural networks has recently attracted a great amount of interest. The many investigations in this area lack a precise common foundation of robustness concepts. Therefore, in this paper, we propose a rigorous and flexible framework for defining different types of robustness properties for classifiers. Our robustness concept is based on postulates that robustness of a classifier should be considered as a property that is independent of accuracy, and that it should be defined in purely mathematical terms without reliance on algorithmic procedures for its measurement. We develop a v…
Virginia Tech（弗吉尼亚理工大学）；Tsinghua University（清华大学）；University of California at Berkeley（加州大学伯克利分校）
Graph convolutional networks (GCNs) are powerful tools for graph-structured data. However, they have been recently shown to be vulnerable to topological attacks. To enhance adversarial robustness, we go beyond spectral graph theory to robust graph theory. By challenging the classical graph Laplacian, we propose a new convolution operator that is provably robust in the spectral domain and is incorporated in the GCN architecture to improve expressivity and interpretability. By extending the original graph to a sequence of graphs, we also propose a robust training paradigm that encourages transfe…
University of Texas at Dallas（德克萨斯大学达拉斯分校）；Purdue University（普渡大学）
We investigate the role of transferability of adversarial attacks in the observed vulnerabilities of Deep Neural Networks (DNNs). We demonstrate that introducing randomness to the DNN models is sufficient to defeat adversarial attacks, given that the adversary does not have an unlimited attack budget. Instead of making one specific DNN model robust to perfect knowledge attacks (a.k.a, white box attacks), creating randomness within an army of DNNs completely eliminates the possibility of perfect knowledge acquisition, resulting in a significantly more robust DNN ensemble against the strongest f…
National Taiwan University（国立台湾大学）；Brookhaven National Laboratory（布鲁克海文国家实验室）；Soochow University（东吴大学）；National Sun Yat-sen University（国立中山大学）
Deep learning (DL) has been applied extensively in a wide range of fields. However, it has been shown that DL models are susceptible to a certain kinds of perturbations called \emph{adversarial attacks}. To fully unlock the power of DL in critical fields such as financial trading, it is necessary to address such issues. In this paper, we present a method of constructing perturbed examples and use these examples to boost the robustness of the model. Our algorithm increases the stability of DL models for candlestick classification with respect to perturbations in the input data.
Deep Neural Networks are vulnerable to adversarial attacks even in settings where the attacker has no direct access to the model being attacked. Such attacks usually rely on the principle of transferability, whereby an attack crafted on a surrogate model tends to transfer to the target model. We show that an ensemble of models with misaligned loss gradients can provide an effective defense against transfer-based attacks. Our key insight is that an adversarial example is less likely to fool multiple models in the ensemble if their loss functions do not increase in a correlated fashion. To this …
Sun Yat-sen University（中山大学）；Emory University（埃默里大学）
Recent studies have shown that Graph Convolutional Networks (GCNs) are vulnerable to adversarial attacks on the graph structure. Although multiple works have been proposed to improve their robustness against such structural adversarial attacks, the reasons for the success of the attacks remain unclear. In this work, we theoretically and empirically demonstrate that structural adversarial examples can be attributed to the non-robust aggregation scheme (i.e., the weighted mean) of GCNs. Specifically, our analysis takes advantage of the breakdown point which can quantitatively measure the robustn…
Improving Adversarial Transferability on Vision-Language Pre-training Models via Surrogate-Specific Bias Correction
School of Cyber Science and Engineering, Southeast University（东南大学网络空间安全学院）；Purple Mountain Laboratories（紫金山实验室）；School of Computer Science and Engineering, Southeast University（东南大学计算机科学与工程学院）
X-ray imagery security screening is essential to maintaining transport security against a varying profile of threat or prohibited items. Particular interest lies in the automatic detection and classification of weapons such as firearms and knives within complex and cluttered X-ray security imagery. Here, we address this problem by exploring various end-to-end object detection Convolutional Neural Network (CNN) architectures. We evaluate several leading variants spanning the Faster R-CNN, Mask R-CNN, and RetinaNet architectures to explore the transferability of such models between varying X-ray…
Rutgers University - School of Engineering（罗格斯大学工程学院）；MIT-IBM Watson AI Lab, IBM Research（IBM研究院麻省理工学院-IBM沃森人工智能实验室）
Deep neural networks (DNNs) are vulnerable to adversarial attack despite their tremendous success in many AI fields. Adversarial attack is a method that causes the intended misclassfication by adding imperceptible perturbations to legitimate inputs. Researchers have developed numerous types of adversarial attack methods. However, from the perspective of practical deployment, these methods suffer from several drawbacks such as long attack generating time, high memory cost, insufficient robustness and low transferability. We propose a Content-aware Adversarial Attack Generator (CAG) to achieve r…
