Sliced Wasserstein adversarial training for improving adversarial robustness

Recently, deep-learning-based models have achieved impressive performance on tasks that were previously considered to be extremely challenging. However, recent works have shown that various deep learning models are susceptible to adversarial data samples. In this paper, we propose the sliced Wassers...

Full description

Saved in:
Bibliographic Details
Published inJournal of ambient intelligence and humanized computing Vol. 15; no. 8; pp. 3229 - 3242
Main Authors Lee, Woojin, Lee, Sungyoon, Kim, Hoki, Lee, Jaewook
Format Journal Article
LanguageEnglish
Published Berlin/Heidelberg Springer Berlin Heidelberg 01.08.2024
Springer Nature B.V
Subjects
Artificial Intelligence
Computational Intelligence
Deep learning
Engineering
Error analysis
Methods
Original Research
Robotics and Automation
User Interfaces and Human Computer Interaction
Sliced Wasserstein Distance
Adversarial defense
Adversarial attack
Adversarial training
Online AccessGet full text

Cover

More Information
Summary:Recently, deep-learning-based models have achieved impressive performance on tasks that were previously considered to be extremely challenging. However, recent works have shown that various deep learning models are susceptible to adversarial data samples. In this paper, we propose the sliced Wasserstein adversarial training method to encourage the logit distributions of clean and adversarial data to be similar to each other. We capture the dissimilarity between two distributions using the Wasserstein metric and then align distributions using an end-to-end training process. We present the theoretical background of the motivation for our study by providing generalization error bounds for adversarial data samples. We performed experiments on three standard datasets and the results demonstrate that our method is more robust against white box attacks compared to previous methods.
ISSN:1868-5137
1868-5145
DOI:10.1007/s12652-024-04791-1