There is a Singularity in the Loss Landscape

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteur principal: Lowell, Mark
Format: Preprint
Publié: 2022
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866914877880139776
author Lowell, Mark
author_facet Lowell, Mark
contents Despite the widespread adoption of neural networks, their training dynamics remain poorly understood. We show experimentally that as the size of the dataset increases, a point forms where the magnitude of the gradient of the loss becomes unbounded. Gradient descent rapidly brings the network close to this singularity in parameter space, and further training takes place near it. This singularity explains a variety of phenomena recently observed in the Hessian of neural network loss functions, such as training on the edge of stability and the concentration of the gradient in a top subspace. Once the network approaches the singularity, the top subspace contributes little to learning, even though it constitutes the majority of the gradient.
format Preprint
id arxiv_https___arxiv_org_abs_2201_06964
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle There is a Singularity in the Loss Landscape
Lowell, Mark
Machine Learning
Artificial Intelligence
Despite the widespread adoption of neural networks, their training dynamics remain poorly understood. We show experimentally that as the size of the dataset increases, a point forms where the magnitude of the gradient of the loss becomes unbounded. Gradient descent rapidly brings the network close to this singularity in parameter space, and further training takes place near it. This singularity explains a variety of phenomena recently observed in the Hessian of neural network loss functions, such as training on the edge of stability and the concentration of the gradient in a top subspace. Once the network approaches the singularity, the top subspace contributes little to learning, even though it constitutes the majority of the gradient.
title There is a Singularity in the Loss Landscape
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2201.06964