Universal characteristics of deep neural network loss surfaces from random matrix theory

Open in new window