Adaptive Soft Contrastive Learning

TL;DR

Adaptive Soft Contrastive Learning (ASCL) enhances self-supervised representation learning by incorporating soft inter-sample relations, transforming the traditional instance discrimination into a multi-instance soft discrimination, leading to improved performance and efficiency.

Contribution

This paper introduces ASCL, a novel plug-in module that adaptively models inter-sample relations in contrastive learning, bridging the gap between instance discrimination and natural sample grouping.

Findings

Achieves state-of-the-art performance on multiple benchmarks.

Improves efficiency of self-supervised learning frameworks.

Effectively models inter-sample relations in visual datasets.

Abstract

Self-supervised learning has recently achieved great success in representation learning without human annotations. The dominant method -- that is contrastive learning, is generally based on instance discrimination tasks, i.e., individual samples are treated as independent categories. However, presuming all the samples are different contradicts the natural grouping of similar samples in common visual datasets, e.g., multiple views of the same dog. To bridge the gap, this paper proposes an adaptive method that introduces soft inter-sample relations, namely Adaptive Soft Contrastive Learning (ASCL). More specifically, ASCL transforms the original instance discrimination task into a multi-instance soft discrimination task, and adaptively introduces inter-sample relations. As an effective and concise plug-in module for existing self-supervised learning frameworks, ASCL achieves the best performance on several benchmarks in terms of both performance and efficiency. Code is available at https://github.com/MrChenFeng/ASCL_ICPR2022.