BMB: Balanced Memory Bank for Imbalanced Semi-supervised Learning

TL;DR

This paper introduces BMB, a semi-supervised learning framework that addresses class imbalance in long-tailed data by maintaining a balanced memory bank and adaptive weighting, significantly improving recognition performance.

Contribution

BMB is a novel semi-supervised approach that uses a class-rebalanced memory bank and adaptive weighting to effectively handle imbalanced data in SSL tasks.

Findings

BMB outperforms state-of-the-art methods by 8.2% on ImageNet127 1% labeled subset.

BMB achieves 4.3% higher accuracy on ImageNet-LT 50% labeled data.

The approach effectively mitigates class imbalance in semi-supervised learning.

Abstract

Exploring a substantial amount of unlabeled data, semi-supervised learning (SSL) boosts the recognition performance when only a limited number of labels are provided. However, traditional methods assume that the data distribution is class-balanced, which is difficult to achieve in reality due to the long-tailed nature of real-world data. While the data imbalance problem has been extensively studied in supervised learning (SL) paradigms, directly transferring existing approaches to SSL is nontrivial, as prior knowledge about data distribution remains unknown in SSL. In light of this, we propose Balanced Memory Bank (BMB), a semi-supervised framework for long-tailed recognition. The core of BMB is an online-updated memory bank that caches historical features with their corresponding pseudo labels, and the memory is also carefully maintained to ensure the data therein are class-rebalanced. Additionally, an adaptive weighting module is introduced to work jointly with the memory bank so as to further re-calibrate the biased training process. We conduct experiments on multiple datasets and demonstrate, among other things, that BMB surpasses state-of-the-art approaches by clear margins, for example 8.2$\%$ on the 1$\%$ labeled subset of ImageNet127 (with a resolution of 64$\times$64) and 4.3$\%$ on the 50$\%$ labeled subset of ImageNet-LT.