CoSSL: Co-Learning of Representation and Classifier for Imbalanced Semi-Supervised Learning

TL;DR

This paper introduces CoSSL, a co-learning framework for imbalanced semi-supervised learning that improves representation and classifier learning, especially under shifted test distributions, achieving state-of-the-art results.

Contribution

Proposes a novel decoupled co-learning framework with Tail-class Feature Enhancement for imbalanced SSL and evaluates under realistic shifted distributions.

Findings

Outperforms existing methods across various shifted test distributions.

Achieves state-of-the-art performance on CIFAR-10, CIFAR-100, ImageNet, and Food-101.

Effective in handling data imbalance in semi-supervised learning scenarios.

Abstract

In this paper, we propose a novel co-learning framework (CoSSL) with decoupled representation learning and classifier learning for imbalanced SSL. To handle the data imbalance, we devise Tail-class Feature Enhancement (TFE) for classifier learning. Furthermore, the current evaluation protocol for imbalanced SSL focuses only on balanced test sets, which has limited practicality in real-world scenarios. Therefore, we further conduct a comprehensive evaluation under various shifted test distributions. In experiments, we show that our approach outperforms other methods over a large range of shifted distributions, achieving state-of-the-art performance on benchmark datasets ranging from CIFAR-10, CIFAR-100, ImageNet, to Food-101. Our code will be made publicly available.