When does gradient descent with logistic loss find interpolating two-layer networks?

MaRDI QIDQ6355291arXivFDO

Authors Niladri S. Chatterji, Philip M. Long, Peter L. Bartlett

Publication date 4 December 2020

Published in arXiv

Mathematics Subject Classification ID

Learning and adaptive systems in artificial intelligence (68T05)

arXiv classification

cs.LGmath.OCstat.ML

Abstract: We study the training of finite-width two-layer smoothed ReLU networks for binary classification using the logistic loss. We show that gradient descent drives the training loss to zero if the initial loss is small enough. When the data satisfies certain cluster and separation conditions and the network is wide enough, we show that one step of gradient descent reduces the loss sufficiently that the first result applies.

This page was built for publication: When does gradient descent with logistic loss find interpolating two-layer networks?

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6355291)