Implicit bias of deep linear networks in the large learning rate phase

Re-assign community