Scholarlink

Deep neural networks used for image classification are highly susceptible to adversarial attacks. The de facto method to increase adversarial robustness is to train neural networks with a mixture of adversarial images and unperturbed images, a method also known as adversarial training. However, this method leads to robust overfitting, where the network primarily learns to recognize one specific type of attack used to generate the images while remaining vulnerable to others after training. This study investigates whether combinations of state of the art data augmentation methods with Stochastic Weight Averaging improve adversarial robustness and diminish adversarial overfitting.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Exploring Data Augmentation and Stochastic Weight Averaging as an Alternative to Adversarial Training of Deep Neural Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Anabetsy Termini,
  • Sumitra Mukherjee

摘要

Deep neural networks used for image classification are highly susceptible to adversarial attacks. The de facto method to increase adversarial robustness is to train neural networks with a mixture of adversarial images and unperturbed images, a method also known as adversarial training. However, this method leads to robust overfitting, where the network primarily learns to recognize one specific type of attack used to generate the images while remaining vulnerable to others after training. This study investigates whether combinations of state of the art data augmentation methods with Stochastic Weight Averaging improve adversarial robustness and diminish adversarial overfitting.