On the Effectiveness of Supervision in Asymmetric Non-Contrastive Learning

TL;DR

This paper explores supervised asymmetric non-contrastive learning (ANCL), demonstrating its effectiveness in improving representation quality and reducing intra-class variance across multiple datasets, with novel supervised frameworks SupSiam and SupBYOL.

Contribution

It introduces supervised ANCL frameworks SupSiam and SupBYOL, extending ANCL to supervised learning and showing their advantages over existing methods.

Findings

Supervised ANCL improves representations and avoids collapse.

Supervision reduces intra-class variance.

Supervised ANCL outperforms other methods across datasets.

Abstract

Supervised contrastive representation learning has been shown to be effective in various transfer learning scenarios. However, while asymmetric non-contrastive learning (ANCL) often outperforms its contrastive learning counterpart in self-supervised representation learning, the extension of ANCL to supervised scenarios is less explored. To bridge the gap, we study ANCL for supervised representation learning, coined SupSiam and SupBYOL, leveraging labels in ANCL to achieve better representations. The proposed supervised ANCL framework improves representation learning while avoiding collapse. Our analysis reveals that providing supervision to ANCL reduces intra-class variance, and the contribution of supervision should be adjusted to achieve the best performance. Experiments demonstrate the superiority of supervised ANCL across various datasets and tasks. The code is available at: https://github.com/JH-Oh-23/Sup-ANCL.