Tweeted By @ml_review

Cross-Entropy Loss Leads To Poor Margins#iclr2019

Show that cross-entropy loss is one of the hidden culprits of adversarial examples
Introduce differential training – novel training paradigm that uses a loss function defined on pairs of from each classhttps://t.co/ognmZEdUOE pic.twitter.com/jslgXF0Xw1

— ML Review (@ml_review) October 14, 2018