On Lazy Training in Differentiable Programming

Abstract : In a series of recent theoretical works, it was shown that strongly over-parameterized neural networks trained with gradient-based methods could converge exponentially fast to zero training loss, with their parameters hardly varying. In this work, we show that this "lazy training" phenomenon is not specific to over-parameterized neural networks, and is due to a choice of scaling, often implicit, that makes the model behave as its linearization around the initialization, thus yielding a model equivalent to learning with positive-definite kernels. Through a theoretical analysis, we exhibit various situations where this phenomenon arises in non-convex optimization and we provide bounds on the distance between the lazy and linearized optimization paths. Our numerical experiments bring a critical note, as we observe that the performance of commonly used non-linear deep convolutional neural networks in computer vision degrades when trained in the lazy regime. This makes it unlikely that "lazy training" is behind the many successes of neural networks in difficult high dimensional tasks.
Document type :
Preprints, Working Papers, ...
Complete list of metadatas

Cited literature [16 references]  Display  Hide  Download

Contributor : Lénaïc Chizat <>
Submitted on : Tuesday, June 18, 2019 - 12:36:13 PM
Last modification on : Friday, June 21, 2019 - 1:31:38 AM


Files produced by the author(s)


  • HAL Id : hal-01945578, version 5
  • ARXIV : 1812.07956


Lenaic Chizat, Edouard Oyallon, Francis Bach. On Lazy Training in Differentiable Programming. 2019. ⟨hal-01945578v5⟩



Record views


Files downloads