long-tailed distribution

The paper presents a novel method called Geometric Harmonization (GH) to address the issue of representation learning disparity in self-supervised learning (SSL), particularly under long-tailed distribution scenarios. Traditional SSL methods…