Proposing SelMix, a selective mixup-based fine-tuning technique for optimizing non-decomposable objectives in machine learning.
