no code implementations • ICLR 2022 • Julian Büchel, Fynn Firouz Faber, Dylan Richard Muir
We present a new network training algorithm that attacks network parameters during training, and promotes robust performance during inference in the face of random parameter variation.
