Geometry and Local Recovery of Global Minima of Two-layer Neural Networks at Overparameterization

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Leyang, Zhang, Yaoyu, Luo, Tao
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908310189375488
author Zhang, Leyang
Zhang, Yaoyu
Luo, Tao
author_facet Zhang, Leyang
Zhang, Yaoyu
Luo, Tao
contents Under mild assumptions, we investigate the geometry of the loss landscape for two-layer neural networks in the vicinity of global minima. Utilizing novel techniques, we demonstrate: (i) how global minima with zero generalization error become geometrically separated from other global minima as the sample size grows; and (ii) the local convergence properties and rate of gradient flow dynamics. Our results indicate that two-layer neural networks can be locally recovered in the regime of overparameterization.
format Preprint
id arxiv_https___arxiv_org_abs_2309_00508
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Geometry and Local Recovery of Global Minima of Two-layer Neural Networks at Overparameterization
Zhang, Leyang
Zhang, Yaoyu
Luo, Tao
Machine Learning
Dynamical Systems
Under mild assumptions, we investigate the geometry of the loss landscape for two-layer neural networks in the vicinity of global minima. Utilizing novel techniques, we demonstrate: (i) how global minima with zero generalization error become geometrically separated from other global minima as the sample size grows; and (ii) the local convergence properties and rate of gradient flow dynamics. Our results indicate that two-layer neural networks can be locally recovered in the regime of overparameterization.
title Geometry and Local Recovery of Global Minima of Two-layer Neural Networks at Overparameterization
topic Machine Learning
Dynamical Systems
url https://arxiv.org/abs/2309.00508