#machine learning
Jul 2026
Class-Balanced Softmax: A Bayes Theory-Based Method for Long-Tailed Recognition
This work characterises a fundamental phenomenon in models trained on imbalanced data, termed the preference issue, wherein models exhibit higher training error and a larger generalisation gap for classes with limited data and demonstrates that CBS effectively mitigates the preference issue.
Yi-Hang Zhu, Rajeev Raman, Shiqi Su et al.
· arXiv.org · 0 citations