Improving Semi-Supervised Contrastive Learning via Entropy-Weighted Confidence Integration of Anchor-Positive Pairs
This is an incremental improvement for semi-supervised learning methods, addressing confidence estimation in pseudo-labeling.
The paper tackles the problem of pseudo-label assignment in semi-supervised contrastive learning by proposing a novel loss function that uses entropy-based confidence weighting to include previously excluded samples, resulting in improved classification accuracy and more stable learning under low-label conditions.
Conventional semi-supervised contrastive learning methods assign pseudo-labels only to samples whose highest predicted class probability exceeds a predefined threshold, and then perform supervised contrastive learning using those selected samples. In this study, we propose a novel loss function that estimates the confidence of each sample based on the entropy of its predicted probability distribution and applies confidence-based adaptive weighting. This approach enables pseudo-label assignment even to samples that were previously excluded from training and facilitates contrastive learning that accounts for the confidence of both anchor and positive samples in a more principled manner. Experimental results demonstrate that the proposed method improves classification accuracy and achieves more stable learning performance even under low-label conditions.