Gradient descent provably escapes saddle points in the training of shallow ReLU networks

Devs

Gradient descent provably escapes saddle points in the training of shallow ReLU networks | Read Paper on Bytez