Provable Stochastic Optimization for Global Contrastive Learning: Small Batch Does Not Harm Performance

Zhuoning Yuan, Yuexin Wu, Zi-Hao Qiu +4

In this paper, we study contrastive learning from an optimization perspective, aiming to analyze and address a fundamental issue of existing contrastive learning methods that eithe…