Has anyone worked on this confidence penalty loss function? What are your experiences with it?
From: Regularizing Neural Networks by Penalizing Confident Output Distributions
Quite promising results, and was wondering if there is an existing implementation in fastai/pytorch.