On Feature Learning in the Presence of Spurious Correlations

TL;DR

This paper investigates how standard ERM and specialized training methods learn core features versus spurious correlations, revealing that simple ERM often learns highly competitive features and that model design choices significantly influence feature quality.

Contribution

The study demonstrates that simple ERM can produce feature representations comparable to specialized methods and highlights the impact of model design and regularization on feature learning.

Findings

ERM features are highly competitive with specialized methods.

Model architecture and pre-training significantly affect feature quality.

Strong regularization is not necessary for high-quality feature representations.

Abstract

Deep classifiers are known to rely on spurious features $\unicode{x2013}$ patterns which are correlated with the target on the training data but not inherently relevant to the learning problem, such as the image backgrounds when classifying the foregrounds. In this paper we evaluate the amount of information about the core (non-spurious) features that can be decoded from the representations learned by standard empirical risk minimization (ERM) and specialized group robustness training. Following recent work on Deep Feature Reweighting (DFR), we evaluate the feature representations by re-training the last layer of the model on a held-out set where the spurious correlation is broken. On multiple vision and NLP problems, we show that the features learned by simple ERM are highly competitive with the features learned by specialized group robustness methods targeted at reducing the effect of spurious correlations. Moreover, we show that the quality of learned feature representations is greatly affected by the design decisions beyond the training method, such as the model architecture and pre-training strategy. On the other hand, we find that strong regularization is not necessary for learning high quality feature representations. Finally, using insights from our analysis, we significantly improve upon the best results reported in the literature on the popular Waterbirds, CelebA hair color prediction and WILDS-FMOW problems, achieving 97%, 92% and 50% worst-group accuracies, respectively.