Predicting the Generalization Gap in Deep Networks with Margin Distributions

Open in new window