“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

In the past few years, it has become increasingly evident that deep neural networks are not resilient enough to withstand adversarial perturbations in input data, leaving them vulnerable to attack. Various authors have proposed strong adversarial attacks for computer vision and Natural Language Processing (NLP) tasks. As a response, many defense mechanisms have also been proposed to prevent these networks from failing. The significance of defending neural networks against adversarial attacks lies in ensuring that the model’s predictions remain unchanged even if the input data is perturbed. Several methods for adversarial defense in NLP have been proposed, catering to different NLP tasks such as text classification, named entity recognition, and natural language inference. Some of these methods not only defend neural networks against adversarial attacks but also act as a regularization mechanism during training, saving the model from overfitting. This survey aims to review the various methods proposed for adversarial defenses in NLP over the past few years by introducing a novel taxonomy. The survey also highlights the fragility of advanced deep neural networks in NLP and the challenges involved in defending them.

https://dl.acm.org/doi/pdf/10.1145/3593042

A Survey of Adversarial Defences and Robustness in NLP

The outstanding performance of deep neural networks has promoted deep learning applications in a broad set of domains. However, the potential risks caused by adversarial samples have hindered the large-scale deployment of deep learning. In these scenarios, adversarial perturbations, imperceptible to human eyes, significantly decrease the model’s final performance. Many papers have been published on adversarial attacks and their countermeasures in the realm of deep learning. Most focus on evasion attacks, where the adversarial examples are found at test time, as opposed to poisoning attacks where poisoned data is inserted into the training data. Further, it is difficult to evaluate the real threat of adversarial attacks or the robustness of a deep learning model, as there are no standard evaluation methods. Hence, with this article, we review the literature to date. Additionally, we attempt to offer the first analysis framework for a systematic understanding of adversarial attacks. The framework is built from the perspective of cybersecurity to provide a lifecycle for adversarial attacks and defenses.

Adversarial Attacks and Defenses in Deep Learning: From a Perspective of Cybersecurity

Despite their overwhelming success on a wide range of applications, convolutional neural networks (CNNs) are widely recognized to be vulnerable to adversarial examples This intriguing phenomenon led to a competition between adversarial attacks and defense techniques So far, adversarial training is the most widely used method for defending against adversarial attacks It has also been extended to defend against universal adversarial perturbations (UAPs) The SOTA universal adversarial training (UAT) method optimizes a single perturbation for all training samples in the mini-batch In this work, we find that a UAP does not attack all classes equally Inspired by this observation, we identify it as the source of the model having unbalanced robustness To this end, we improve the SOTA UAT by proposing to utilize class-wise UAPs during adversarial training On multiple benchmark datasets, our class-wise UAT leads superior performance for both clean accuracy and adversarial robustness against universal attack

Universal Adversarial Training with Class-Wise Perturbations

In the last decade, deep neural networks have proven to be very powerful in computer vision tasks, starting a revolution in the computer vision and machine learning fields. However, deep neural networks, usually, are not robust to perturbations of the input data. In fact, several studies showed that slightly changing the content of the images can cause a dramatic decrease in the accuracy of the attacked neural network. Several methods able to generate adversarial samples make use of gradients, which usually are not available to an attacker in real-world scenarios. As opposed to this class of attacks, another class of adversarial attacks, called black-box adversarial attacks, emerged, which does not make use of information on the gradients, being more suitable for real-world attack scenarios. In this work, we compare three well-known evolution strategies on the generation of black-box adversarial attacks for image classification tasks. While our results show that the attacked neural networks can be, in most cases, easily fooled by all the algorithms under comparison, they also show that some black-box optimization algorithms may be better in "harder" setups, both in terms of attack success rate and efficiency (i.e., number of queries).

Black-box adversarial attacks using Evolution Strategies

Deep neural networks (DNNs) have demonstrated remarkable performance for various applications, meanwhile, they are widely known to be vulnerable to the attack of adversarial perturbations. This intriguing phenomenon has attracted significant attention in machine learning and what might be more surprising to the community is the existence of universal adversarial perturbations (UAPs), i.e. a single perturbation to fool the target DNN for most images. The advantage of UAP is that it can be generated beforehand and then be applied on-the-fly during the attack. With the focus on UAP against deep classifiers, this survey summarizes the recent progress on universal adversarial attacks, discussing the challenges from both the attack and defense sides, as well as the reason for the existence of UAP. Additionally, universal attacks in a wide range of applications beyond deep classification are also covered.

/pdf/a-survey-on-universal-adversarial-attack-5d631eutyd.pdf

A Survey on Universal Adversarial Attack.

Data hiding is the art of concealing messages with limited perceptual changes. Recently, deep learning has provided enriching perspectives for it and made significant progress. In this work, we conduct a brief yet comprehensive review of existing literature and outline three meta-architectures. Based on this, we summarize specific strategies for various applications of deep hiding, including steganography, light field messaging and watermarking. Finally, further insight into deep hiding is provided through incorporating the perspective of adversarial attack.

A Brief Survey on Deep Learning Based Data Hiding, Steganography and Watermarking.

Digital watermarking has been widely used to protect the copyright and integrity of multimedia data. Previous studies mainly focus on designing watermarking techniques that are robust to attacks of destroying the embedded watermarks. However, the emerging deep learning based image generation technology raises new open issues that whether it is possible to generate fake watermarked images for circumvention. In this paper, we make the first attempt to develop digital image watermark fakers by using generative adversarial learning. Suppose that a set of paired images of original and watermarked images generated by the targeted watermarker are available, we use them to train a watermark faker with U-Net as the backbone, whose input is an original image, and after a domain-specific preprocessing, it outputs a fake watermarked image. Our experiments show that the proposed watermark faker can effectively crack digital image watermarkers in both spatial and frequency domains, suggesting the risk of such forgery attacks.

Chenguo Lin

Papers

A Survey on Universal Adversarial Attack.

A Brief Survey on Deep Learning Based Data Hiding, Steganography and Watermarking.

Watermark Faker: Towards Forgery of Digital Image Watermarking

A Survey On Universal Adversarial Attack

Watermark Faker: Towards Forgery of Digital Image Watermarking