BaCon method addresses imbalanced data distribution in Class Imbalanced Semi-supervised Learning (CISSL).
Existing methods focus on instance-level adjustments, but BaCon emphasizes balanced feature distribution.
BaCon uses contrastive learning to regularize feature representations and achieve better performance.
The method is effective across various datasets like CIFAR10-LT, CIFAR100-LT, STL10-LT, and SVHN-LT.
BaCon outperforms other methods like FixMatch-based ABC and CoSSL in accuracy improvements.
The paper discusses the limitations of existing CISSL methods and proposes a novel approach for more balanced representation learning.
"BaCon directly regularizes the distribution of instances’ representations in a well-designed contrastive manner."
"Our method demonstrates its effectiveness through comprehensive experiments on the CIFAR10-LT, CIFAR100-LT, STL10-LT, and SVHN-LT datasets."
"When encountering more extreme imbalance degree, BaCon also shows better robustness than other methods."