Advances in Adversarial Attacks and Defenses in Computer Vision: A Survey

Akhtar, Naveed; Mian, Ajmal; Kardan, Navid; Shah, Mubarak

doi:10.1109/access.2021.3127960

Cited by 184 publications

(67 citation statements)

References 249 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Since the discovery of adversarial examples, numerous efforts have been made to design advanced adversarial attack methods [10]. Nevertheless, most of the previous research focuses on the white-box attack, where it is assumed that complete knowledge of the victim model, including its architecture and parameter values, is accessible for the attacker.…”

Section: B Black-box Attacksmentioning

confidence: 99%

“…Although these white-box attack methods can achieve high success rates in the attack, their assumption is usually invalid in practical applications. To make the attack more pragmatic, black-box attack methods are proposed which assume no (or minimal) knowledge about the victim model [10]. Generally, black-box attacks can be categorized into query-based and transfer-based methods.…”

Section: B Black-box Attacksmentioning

confidence: 99%

“…For example, Liu et these methods is to change the pixel values of the input image toward the direction that the surrogate model could yield wrong predicted logits. However, due to the discrepancy between different neural networks, the direction that is harmful to the surrogate model may have limited influence on the victim model [10]. To make the adversarial attack more transferable, we consider attacking the features in the shallow layers of the surrogate model.…”

Section: B Black-box Attacksmentioning

confidence: 99%

“…Recall that our goal is to produce adversarial examples that share similar feature representation with the virtual sample x from (9) or (10). To this end, we propose the mix loss L mix , which minimizes the distribution discrepancy between the features from the adversarial example and the virtual samples.…”

Section: Feature-level Attack With Virtual Samplesmentioning

confidence: 99%

“…However, in real-world scenarios, it is usually impossible to obtain detailed information about the deployed network [20]. Under such a circumstance, it is more feasible to conduct a black-box attack, where the adversarial examples are generated without any knowledge of the victim model [10]. Obviously, it is more challenging to implement black-box attacks [21].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Universal Adversarial Examples in Remote Sensing: Methodology and Benchmark

Ghamisi

2022

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

Deep neural networks have achieved great success in many important remote sensing tasks. Nevertheless, their vulnerability to adversarial examples should not be neglected. In this study, we systematically analyze the universal adversarial examples in remote sensing data for the first time, without any knowledge from the victim model. Specifically, we propose a novel black-box adversarial attack method, namely Mixup-Attack, and its simple variant Mixcut-Attack, for remote sensing data. The key idea of the proposed methods is to find common vulnerabilities among different networks by attacking the features in the shallow layer of a given surrogate model. Despite their simplicity, the proposed methods can generate transferable adversarial examples that deceive most of the state-of-the-art deep neural networks in both scene classification and semantic segmentation tasks with high success rates. We further provide the generated universal adversarial examples in the dataset named UAE-RS, which is the first dataset that provides black-box adversarial samples in the remote sensing field. We hope UAE-RS may serve as a benchmark that helps researchers to design deep neural networks with strong resistance toward adversarial attacks in the remote sensing field. Codes and the UAE-RS dataset will be available online 1 .

show abstract

Section: B Black-box Attacksmentioning

confidence: 99%

Section: B Black-box Attacksmentioning

confidence: 99%

Section: B Black-box Attacksmentioning

confidence: 99%

Section: Feature-level Attack With Virtual Samplesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations