Gradient descent with identity initialization efficiently learns positive definite linear transformations by deep residual networks

Add code
Jun 18, 2018

Share this with someone who'll enjoy it:

View paper onarxiv icon

Share this with someone who'll enjoy it: