Vincent Nguyen

SAMix: Calibrated and Accurate Continual Learning via Sphere-Adaptive Mixup and Neural Collapse

Trung-Anh Dang, Vincent Nguyen, Ngoc-Son Vu, Christel Vrain

Abstract While most continual learning methods focus on mitigating forgetting and improving accuracy, they often overlook the critical aspect of network calibration, despite its importance. Neural collapse, a phenomenon where last-layer features collapse to their class means, has…Read more