A provable control of sensitivity of neural networks through a direct parameterization of the overall bi-Lipschitzness

arXiv:2404.09821v1 Announce Type: new
Abstract: While neural networks can enjoy an outstanding flexibility and exhibit unprecedented performance, the mechanism behind their behavior is still not well-understood. To tackle this fundamental challenge, researchers have tried to restrict and manipulate some of their properties in order to gain new insights and better control on them. Especially, throughout the past few years, the concept of \emph{bi-Lipschitzness} has been proved as a beneficial inductive bias in many areas. However, due to its complexity, the …

abstract arxiv behavior challenge control cs.lg flexibility networks neural networks performance researchers sensitivity stat.ml through type