Imbalanced Semi Supervised Learning

Imbalanced semi-supervised learning (SSL) tackles the challenge of training machine learning models with limited labeled data and a skewed class distribution in both labeled and unlabeled datasets. Current research focuses on mitigating the bias introduced by unreliable pseudo-labels generated from the imbalanced data, employing techniques like class-distribution-aware debiasing, balanced contrastive learning, and refined pseudo-label generation strategies. These advancements aim to improve model accuracy and robustness in real-world applications where obtaining sufficient labeled data is costly or impractical, impacting fields like image classification and point cloud segmentation.