Tradeoffs Between Contrastive and Supervised Learning: An Empirical Study

TL;DR

This paper empirically compares contrastive and supervised learning, revealing that supervised pretraining can outperform contrastive methods under limited training budgets and on certain tasks, emphasizing the importance of context in choosing pretraining strategies.

Contribution

It provides a comprehensive empirical analysis of the tradeoffs between contrastive and supervised learning across different training budgets and tasks, highlighting scenarios where supervised learning excels.

Findings

Supervised pretraining outperforms contrastive learning with limited compute budgets.

Supervised learning remains more effective on certain tasks despite larger pretraining budgets.

Object-centric bias in supervised pretraining enhances robustness to corruptions.

Abstract

Contrastive learning has made considerable progress in computer vision, outperforming supervised pretraining on a range of downstream datasets. However, is contrastive learning the better choice in all situations? We demonstrate two cases where it is not. First, under sufficiently small pretraining budgets, supervised pretraining on ImageNet consistently outperforms a comparable contrastive model on eight diverse image classification datasets. This suggests that the common practice of comparing pretraining approaches at hundreds or thousands of epochs may not produce actionable insights for those with more limited compute budgets. Second, even with larger pretraining budgets we identify tasks where supervised learning prevails, perhaps because the object-centric bias of supervised pretraining makes the model more resilient to common corruptions and spurious foreground-background correlations. These results underscore the need to characterize tradeoffs of different pretraining objectives across a wider range of contexts and training regimes.