Abstract

One of the main difficulties in analyzing neural networks is thenon-convexity of the loss function which may have many bad local minima. In this paper, we study the landscape of neural networks for binaryclassification tasks. Under mild assumptions, we prove that after adding onespecial neuron with a skip connection to the output, or one special neuron perlayer, every local minimum is a global minimum.

 

Quick Read (beta)