1 paper across 1 session
We introduce a regularizer that promotes diverse, task-relevant features over spurious ones in contrastive learning.