THE IMPLICIT BIAS OF DEPTH: HOW INCREMENTAL LEARNING DRIVES GENERALIZATION

Research output: Contribution to conferencePaperpeer-review

18 Scopus citations

Abstract

A leading hypothesis for the surprising generalization of neural networks is that the dynamics of gradient descent bias the model towards simple solutions, by searching through the solution space in an incremental order of complexity. We formally define the notion of incremental learning dynamics and derive the conditions on depth and initialization for which this phenomenon arises in deep linear models. Our main theoretical contribution is a dynamical depth separation result, proving that while shallow models can exhibit incremental learning dynamics, they require the initialization to be exponentially small for these dynamics to present themselves. However, once the model becomes deeper, the dependence becomes polynomial and incremental learning can arise in more natural settings. We complement our theoretical findings by experimenting with deep matrix sensing, quadratic neural networks and with binary classification using diagonal and convolutional linear networks, showing all of these models exhibit incremental learning.

Original languageAmerican English
StatePublished - 2020
Event8th International Conference on Learning Representations, ICLR 2020 - Addis Ababa, Ethiopia
Duration: 30 Apr 2020 → …

Conference

Conference8th International Conference on Learning Representations, ICLR 2020
Country/TerritoryEthiopia
CityAddis Ababa
Period30/04/20 → …

Bibliographical note

Funding Information:
This research is supported by the European Research Council (TheoryDL project).

Publisher Copyright:
© 2020 8th International Conference on Learning Representations, ICLR 2020. All rights reserved.

Fingerprint

Dive into the research topics of 'THE IMPLICIT BIAS OF DEPTH: HOW INCREMENTAL LEARNING DRIVES GENERALIZATION'. Together they form a unique fingerprint.

Cite this