Limitations of Post-Hoc Feature Alignment for Robustness

TL;DR

This paper critically examines the effectiveness of post-hoc feature alignment, specifically batch normalization statistics matching, revealing its limited benefits and potential drawbacks for robustness against distribution shifts.

Contribution

The study provides a detailed analysis of the limitations of feature alignment methods, especially batch normalization alignment, and explains why they may not reliably improve robustness.

Findings

Only helps with specific distribution shifts

Can degrade performance in some settings

Challenges the utility of feature alignment for robustness

Abstract

Feature alignment is an approach to improving robustness to distribution shift that matches the distribution of feature activations between the training distribution and test distribution. A particularly simple but effective approach to feature alignment involves aligning the batch normalization statistics between the two distributions in a trained neural network. This technique has received renewed interest lately because of its impressive performance on robustness benchmarks. However, when and why this method works is not well understood. We investigate the approach in more detail and identify several limitations. We show that it only significantly helps with a narrow set of distribution shifts and we identify several settings in which it even degrades performance. We also explain why these limitations arise by pinpointing why this approach can be so effective in the first place. Our findings call into question the utility of this approach and Unsupervised Domain Adaptation more broadly for improving robustness in practice.