ATVis: Understanding and diagnosing adversarial training processes through visual analytics

Adversarial training has emerged as a major strategy against adversarial perturbations in deep neural networks, which mitigates the issue of exploiting model vulnerabilities to generate incorrect predictions. Despite enhancing robustness, adversarial training often results in a trade-off with standa...

Full description

Saved in:
Bibliographic Details
Main Authors: Fang Zhu, Xufei Zhu, Xumeng Wang, Yuxin Ma, Jieqiong Zhao
Format: Article
Language:English
Published: Elsevier 2024-12-01
Series:Visual Informatics
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2468502X24000639
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850059006290165760
author Fang Zhu
Xufei Zhu
Xumeng Wang
Yuxin Ma
Jieqiong Zhao
author_facet Fang Zhu
Xufei Zhu
Xumeng Wang
Yuxin Ma
Jieqiong Zhao
author_sort Fang Zhu
collection DOAJ
description Adversarial training has emerged as a major strategy against adversarial perturbations in deep neural networks, which mitigates the issue of exploiting model vulnerabilities to generate incorrect predictions. Despite enhancing robustness, adversarial training often results in a trade-off with standard accuracy on normal data, a phenomenon that remains a contentious issue. In addition, the opaque nature of deep neural network models renders it more difficult to inspect and diagnose how adversarial training processes evolve. This paper introduces ATVis, a visual analytics framework for examining and diagnosing adversarial training processes. Through multi-level visualization design, ATVis enables the examination of model robustness from various granularity, facilitating a detailed understanding of the dynamics in the training epochs. The framework reveals the complex relationship between adversarial robustness and standard accuracy, which further offers insights into the mechanisms that drive the trade-offs observed in adversarial training. The effectiveness of the framework is demonstrated through case studies.
format Article
id doaj-art-a0404862fe2141f4ba7cbda8f87a685b
institution DOAJ
issn 2468-502X
language English
publishDate 2024-12-01
publisher Elsevier
record_format Article
series Visual Informatics
spelling doaj-art-a0404862fe2141f4ba7cbda8f87a685b2025-08-20T02:51:00ZengElsevierVisual Informatics2468-502X2024-12-0184718410.1016/j.visinf.2024.10.003ATVis: Understanding and diagnosing adversarial training processes through visual analyticsFang Zhu0Xufei Zhu1Xumeng Wang2Yuxin Ma3Jieqiong Zhao4Southern University of Science and Technology, Shenzhen, ChinaSouthern University of Science and Technology, Shenzhen, ChinaDISSec, Nankai University, Tianjin, China; Corresponding authors.Southern University of Science and Technology, Shenzhen, China; Corresponding authors.Augusta University, Augusta, Georgia, USAAdversarial training has emerged as a major strategy against adversarial perturbations in deep neural networks, which mitigates the issue of exploiting model vulnerabilities to generate incorrect predictions. Despite enhancing robustness, adversarial training often results in a trade-off with standard accuracy on normal data, a phenomenon that remains a contentious issue. In addition, the opaque nature of deep neural network models renders it more difficult to inspect and diagnose how adversarial training processes evolve. This paper introduces ATVis, a visual analytics framework for examining and diagnosing adversarial training processes. Through multi-level visualization design, ATVis enables the examination of model robustness from various granularity, facilitating a detailed understanding of the dynamics in the training epochs. The framework reveals the complex relationship between adversarial robustness and standard accuracy, which further offers insights into the mechanisms that drive the trade-offs observed in adversarial training. The effectiveness of the framework is demonstrated through case studies.http://www.sciencedirect.com/science/article/pii/S2468502X24000639Visual analyticsExplainable AIAdversarial training
spellingShingle Fang Zhu
Xufei Zhu
Xumeng Wang
Yuxin Ma
Jieqiong Zhao
ATVis: Understanding and diagnosing adversarial training processes through visual analytics
Visual Informatics
Visual analytics
Explainable AI
Adversarial training
title ATVis: Understanding and diagnosing adversarial training processes through visual analytics
title_full ATVis: Understanding and diagnosing adversarial training processes through visual analytics
title_fullStr ATVis: Understanding and diagnosing adversarial training processes through visual analytics
title_full_unstemmed ATVis: Understanding and diagnosing adversarial training processes through visual analytics
title_short ATVis: Understanding and diagnosing adversarial training processes through visual analytics
title_sort atvis understanding and diagnosing adversarial training processes through visual analytics
topic Visual analytics
Explainable AI
Adversarial training
url http://www.sciencedirect.com/science/article/pii/S2468502X24000639
work_keys_str_mv AT fangzhu atvisunderstandinganddiagnosingadversarialtrainingprocessesthroughvisualanalytics
AT xufeizhu atvisunderstandinganddiagnosingadversarialtrainingprocessesthroughvisualanalytics
AT xumengwang atvisunderstandinganddiagnosingadversarialtrainingprocessesthroughvisualanalytics
AT yuxinma atvisunderstandinganddiagnosingadversarialtrainingprocessesthroughvisualanalytics
AT jieqiongzhao atvisunderstandinganddiagnosingadversarialtrainingprocessesthroughvisualanalytics