Almost-Orthogonal Layers for Efficient General-Purpose Lipschitz Networks

Published in ECCV, 2022

We introduce an efficient rescaling-based layer that allows us to train state-of-the-art certifiably robust image classifiers.