Standard-Deviation-Inspired Regularization for Improving Adversarial Robustness

TL;DR

This paper introduces a novel regularization method inspired by standard deviation to enhance adversarial robustness and generalization in deep neural networks, complementing existing adversarial training techniques.

Contribution

It proposes a standard-deviation-inspired regularization term that improves robustness and generalization when combined with adversarial training, supported by experimental validation.

Findings

SDI measure can craft adversarial examples

Combining SDI with AT enhances robustness against strong attacks

SDI regularization improves model generalization

Abstract

Adversarial Training (AT) has been demonstrated to improve the robustness of deep neural networks (DNNs) against adversarial attacks. AT is a min-max optimization procedure where in adversarial examples are generated to train a more robust DNN. The inner maximization step of AT increases the losses of inputs with respect to their actual classes. The outer minimization involves minimizing the losses on the adversarial examples obtained from the inner maximization. This work proposes a standard-deviation-inspired (SDI) regularization term to improve adversarial robustness and generalization. We argue that the inner maximization in AT is similar to minimizing a modified standard deviation of the model's output probabilities. Moreover, we suggest that maximizing this modified standard deviation can complement the outer minimization of the AT framework. To support our argument, we experimentally show that the SDI measure can be used to craft adversarial examples. Additionally, we demonstrate that combining the SDI regularization term with existing AT variants enhances the robustness of DNNs against stronger attacks, such as CW and Auto-attack, and improves generalization.