Difficult Examples Hurt Unsupervised Contrastive Learning: A Theoretical Perspective

TL;DR

This paper provides a theoretical analysis showing that removing difficult examples can improve unsupervised contrastive learning's generalization and performance, challenging previous assumptions about their importance.

Contribution

It develops a theoretical framework explaining how difficult examples negatively impact contrastive learning and proposes methods to improve it by removing or tuning these examples.

Findings

Removing difficult examples boosts downstream classification performance.

Techniques like margin tuning and temperature scaling enhance generalization bounds.

Empirical validation confirms the effectiveness of the proposed methods.

Abstract

Unsupervised contrastive learning has shown significant performance improvements in recent years, often approaching or even rivaling supervised learning in various tasks. However, its learning mechanism is fundamentally different from supervised learning. Previous works have shown that difficult examples (well-recognized in supervised learning as examples around the decision boundary), which are essential in supervised learning, contribute minimally in unsupervised settings. In this paper, perhaps surprisingly, we find that the direct removal of difficult examples, although reduces the sample size, can boost the downstream classification performance of contrastive learning. To uncover the reasons behind this, we develop a theoretical framework modeling the similarity between different pairs of samples. Guided by this framework, we conduct a thorough theoretical analysis revealing that the presence of difficult examples negatively affects the generalization of contrastive learning. Furthermore, we demonstrate that the removal of these examples, and techniques such as margin tuning and temperature scaling can enhance its generalization bounds, thereby improving performance. Empirically, we propose a simple and efficient mechanism for selecting difficult examples and validate the effectiveness of the aforementioned methods, which substantiates the reliability of our proposed theoretical framework.