Learning with a Strong Adversary

TL;DR

This paper introduces a new training method that enhances neural network robustness by incorporating adversarial example generation during learning, leading to more resilient classifiers.

Contribution

It presents a novel approach called learning with a strong adversary, which integrates efficient adversarial example generation into the training process to improve robustness.

Findings

Significantly improved robustness of classifiers.

Efficient method for generating adversarial examples.

Experimental validation demonstrating robustness gains.

Abstract

The robustness of neural networks to intended perturbations has recently attracted significant attention. In this paper, we propose a new method, \emph{learning with a strong adversary}, that learns robust classifiers from supervised data. The proposed method takes finding adversarial examples as an intermediate step. A new and simple way of finding adversarial examples is presented and experimentally shown to be efficient. Experimental results demonstrate that resulting learning method greatly improves the robustness of the classification models produced.