Decoder-free Robustness Disentanglement without (Additional) Supervision

TL;DR

This paper introduces a novel adversarial training method that disentangles robust and non-robust features without extra supervision, improving model robustness and interpretability.

Contribution

The proposed Adversarial Asymmetric Training (AAT) method effectively separates robust and non-robust features without additional labels, enhancing disentanglement and accuracy.

Findings

Successfully preserves accuracy by combining two representations

Achieves better disentanglement than previous methods

No additional supervision needed for robustness separation

Abstract

Adversarial Training (AT) is proposed to alleviate the adversarial vulnerability of machine learning models by extracting only robust features from the input, which, however, inevitably leads to severe accuracy reduction as it discards the non-robust yet useful features. This motivates us to preserve both robust and non-robust features and separate them with disentangled representation learning. Our proposed Adversarial Asymmetric Training (AAT) algorithm can reliably disentangle robust and non-robust representations without additional supervision on robustness. Empirical results show our method does not only successfully preserve accuracy by combining two representations, but also achieve much better disentanglement than previous work.