Generating Less Certain Adversarial Examples Improves Robust Generalization

TL;DR

This paper shows that generating less certain adversarial examples during training enhances a model's robustness and reduces overfitting, by formalizing adversarial certainty and developing a method to produce such examples.

Contribution

It introduces a formal measure of adversarial certainty, links it to robust generalization, and proposes a method to generate less certain adversarial inputs for improved robustness.

Findings

Models trained with less certain adversarial examples show better robustness.

The proposed method mitigates robust overfitting in image classification.

Experimental results confirm improved robustness across benchmarks.

Abstract

This paper revisits the robust overfitting phenomenon of adversarial training. Observing that models with better robust generalization performance are less certain in predicting adversarially generated training inputs, we argue that overconfidence in predicting adversarial examples is a potential cause. Therefore, we hypothesize that generating less certain adversarial examples improves robust generalization, and propose a formal definition of adversarial certainty that captures the variance of the model's predicted logits on adversarial examples. Our theoretical analysis of synthetic distributions characterizes the connection between adversarial certainty and robust generalization. Accordingly, built upon the notion of adversarial certainty, we develop a general method to search for models that can generate training-time adversarial inputs with reduced certainty, while maintaining the model's capability in distinguishing adversarial examples. Extensive experiments on image benchmarks demonstrate that our method effectively learns models with consistently improved robustness and mitigates robust overfitting, confirming the importance of generating less certain adversarial examples for robust generalization. Our implementations are available as open-source code at: https://github.com/TrustMLRG/AdvCertainty.