Class-Imbalanced Semi-Supervised Learning with Inverse Auxiliary Classifier

Tiansong Jiang (Nanjing University of Science and Technology),* Sheng Wan (Nanjing university of science and technology), Chen Gong (Nanjing University of Science and Technology)
The 34th British Machine Vision Conference


In this paper, we focus on the challenge posed by Class-Imbalanced Semi-Supervised Learning (CISSL). Existing pseudo-labeling-based Semi-Supervised Learning (SSL) algorithms often exhibit poor performance in minority classes, thereby resulting in degradation of the feature learning process. This issue becomes more pronounced when labeled and unlabeled data exhibit different class distributions. To mitigate the effect of imbalanced labeled data on feature learning, we introduce a simple yet effective plug-in module, i.e., Inverse Auxiliary Classifier (IAC). The module utilizes a down-sampling strategy by using a mask that inverts the class distribution of labeled data. Additionally, we propose an Inverse Distribution Alignment (IDA) loss to encourage IAC to focus on the underrepresented minority classes in labeled data. The proposed method can be seamlessly integrated into multiple existing CISSL algorithms without any difficulty. Extensive experiments conducted in this paper demonstrate that incorporating the proposed IAC can improve the performance of different CISSL models, especially when there is a significant disparity between the class distributions of labeled and unlabeled data.



author    = {Tiansong Jiang and Sheng Wan and Chen Gong},
title     = {Class-Imbalanced Semi-Supervised Learning with Inverse Auxiliary Classifier},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year      = {2023},
url       = {}

Copyright © 2023 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection