Difficult Examples Hurt Unsupervised Contrastive Learning: A Theoretical Perspective

Yi-Ge Zhang, Jingyi Cui, Qiran Li, Yisen Wang

Keywords: Machine Learning. Self-Supervised Learning. Difficult Examples

TL;DR: We introduce a similarity-based theoretical framework that shows how difficult boundary examples impair generalization in unsupervised contrastive learning, and we design mechanisms that address this issue and boost downstream accuracy.

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.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Submission Number: 3401