Deep Neural Nets as Hamiltonians

Mike Winer, Boris Hanin

Neural networks are complex functions of both their inputs and parameters. Much prior work in deep learning theory analyzes the distribution of network outputs at a fixed a set of…