Skip to content

Latest commit

 

History

History
3 lines (3 loc) · 625 Bytes

File metadata and controls

3 lines (3 loc) · 625 Bytes

Truth or backpropaganda? An empirical investigation of deep learning theory (https://openreview.net/forum?id=HyxyIgHFvr)

Micah Goldblum, Jonas Geiping, Avi Schwarzschild, Michael Moeller, Tom Goldstein

This repository consists of code developed in order to empirically investigate assertions from deep learning theory. The associated paper can be found here. Specifically, experiments in this repository examine local minima of the loss surface, weight decay, neural tangent kernels, and the rank of operators in networks. The directories in this repository are separated by topic.