Adversarial Training with Orthogonal Regularization

Yüksel, Oğuz Kaan; Baytaş, İnci M.

doi:10.1109/siu49456.2020.9302247

Cited by 2 publications

(1 citation statement)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As a result, PGD adversarial training is one of the most effective adversarial defense techniques. On the other hand, the PGD adversarial training has a poor generalization performance due to the lack of diversity in the PGD attacks [22]. There is a growing number of studies that investigate more robust and more generalizable adversarial defense techniques.…”

Section: Related Workmentioning

confidence: 99%

Robustness-via-Synthesis: Robust Training with Generative Adversarial Perturbations

Baytaş¹,

Deb²

2021

Preprint

View full text Add to dashboard Cite

Upon the discovery of adversarial attacks, robust models have become obligatory for deep learning-based systems.Adversarial training with first-order attacks has been one of the most effective defenses against adversarial perturbations to this day. The majority of the adversarial training approaches focus on iteratively perturbing each pixel with the gradient of the loss function with respect to the input image. However, the adversarial training with gradient-based attacks lacks diversity and does not generalize well to natural images and various attacks. This study presents a robust training algorithm where the adversarial perturbations are automatically synthesized from a random vector using a generator network. The classifier is trained with cross-entropy loss regularized with the optimal transport distance between the representations of the natural and synthesized adversarial samples. Unlike prevailing generative defenses, the proposed one-step attack generation framework synthesizes diverse perturbations without utilizing gradient of the classifier's loss. Experimental results show that the proposed approach attains comparable robustness with various gradientbased and generative robust training techniques on CIFAR10, CIFAR100, and SVHN datasets. In addition, compared to the baselines, the proposed robust training framework generalizes well to the natural samples. Code and trained models will be made publicly available.

show abstract

Section: Related Workmentioning

confidence: 99%