Data-Efficient Contrastive Self-supervised Learning: Most Beneficial Examples for Supervised Learning Contribute the Least

TL;DR

This paper identifies which examples are most beneficial for contrastive self-supervised learning, showing that examples with highly similar augmentations to others are most valuable, enabling data reduction without performance loss.

Contribution

It provides the first theoretical analysis linking example similarity to their contribution in contrastive SSL and demonstrates effective data subset selection for efficiency.

Findings

Excluding 20-40% of data does not harm downstream performance.

Selected subsets outperform random subsets by over 3%.

Highly similar augmentation examples contribute most to contrastive SSL.

Abstract

Self-supervised learning (SSL) learns high-quality representations from large pools of unlabeled training data. As datasets grow larger, it becomes crucial to identify the examples that contribute the most to learning such representations. This enables efficient SSL by reducing the volume of data required. Nevertheless, quantifying the value of examples for SSL has remained an open question. In this work, we address this problem for the first time, by proving that examples that contribute the most to contrastive SSL are those that have the most similar augmentations to other examples, in expectation. We provide rigorous guarantees for the generalization performance of contrastive learning on such subsets. Through extensive experiments, we show that we can safely exclude 20% of examples from CIFAR100 and 40% from STL10 and TinyImageNet, without affecting downstream task performance. In general, subsets selected by our method outperform random subsets by over 3% across these datasets. Interestingly, we also discover the subsets that contribute the most to contrastive learning are those that contribute the least to supervised learning. Code available at https://github.com/bigml-cs-ucla/sas-data-efficient-contrastive-learning.