> what it is doing is applying a probabilistic best-fit I think you're underselling probabalistic best-fits. Especially with all of the regularization going on

> what it is doing is applying a probabilistic best-fit I think you're underselling probabalistic best-fits. Especially with all of the regularization going on in training.