36ad8b5f42db492827016448975cc22d-AuthorFeedback.pdf

Note that mixup models can turn out to beunderconfidentifα is large. In our experiments, we see the resulting models are prone to under-fitting and28 under-confidence.