Fetching the paper…

Neural networks trained with SGD learn distributions of increasing complexity · Around