Skip to Main content Skip to Navigation
Conference papers

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 :
Conference papers
Complete list of metadatas

Cited literature [16 references]  Display  Hide  Download

https://hal.inria.fr/hal-01945578
Contributor : Lénaïc Chizat <>
Submitted on : Tuesday, January 7, 2020 - 4:58:33 PM
Last modification on : Wednesday, November 18, 2020 - 6:21:03 PM

Files

lazy-main.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01945578, version 6
  • ARXIV : 1812.07956

Collections

Citation

Lenaic Chizat, Edouard Oyallon, Francis Bach. On Lazy Training in Differentiable Programming. NeurIPS 2019 - 33rd Conference on Neural Information Processing Systems, Dec 2019, Vancouver, Canada. pp.2937-2947. ⟨hal-01945578v6⟩

Share

Metrics

Record views

11137

Files downloads

605