Bayesian Inference with Shaped Deep Non-linear MLPs

Boris Hanin, Tianze Jiang

A central aim of deep learning theory is to characterize how neural networks make predictions in the regime of simultaneously large model and training set size. Since the limits of…