Which Features are Learnt by Contrastive Learning? On the Role of Simplicity Bias in Class Collapse and Feature Suppression

TL;DR

This paper provides a theoretical framework explaining how contrastive learning tends to favor simpler features, leading to class collapse and feature suppression, and proposes solutions like increasing embedding dimensionality and better data augmentations.

Contribution

It offers the first rigorous theory on feature learning in contrastive learning, linking simplicity bias to class collapse and feature suppression, and suggests practical improvements.

Findings

Gradient bias towards simplicity causes class collapse.

Increasing embedding size mitigates feature suppression.

Better data augmentations improve feature learning.

Abstract

Contrastive learning (CL) has emerged as a powerful technique for representation learning, with or without label supervision. However, supervised CL is prone to collapsing representations of subclasses within a class by not capturing all their features, and unsupervised CL may suppress harder class-relevant features by focusing on learning easy class-irrelevant features; both significantly compromise representation quality. Yet, there is no theoretical understanding of \textit{class collapse} or \textit{feature suppression} at \textit{test} time. We provide the first unified theoretically rigorous framework to determine \textit{which} features are learnt by CL. Our analysis indicate that, perhaps surprisingly, bias of (stochastic) gradient descent towards finding simpler solutions is a key factor in collapsing subclass representations and suppressing harder class-relevant features. Moreover, we present increasing embedding dimensionality and improving the quality of data augmentations as two theoretically motivated solutions to {feature suppression}. We also provide the first theoretical explanation for why employing supervised and unsupervised CL together yields higher-quality representations, even when using commonly-used stochastic gradient methods.