Understanding Two-Layer Neural Networks with Smooth Activation Functions

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autor principal: Huang, Changcun
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866913948806152192
author Huang, Changcun
author_facet Huang, Changcun
contents This paper aims to understand the training solution, which is obtained by the back-propagation algorithm, of two-layer neural networks whose hidden layer is composed of the units with smooth activation functions, including the usual sigmoid type most commonly used before the advent of ReLUs. The mechanism contains four main principles: construction of Taylor series expansions, strict partial order of knots, smooth-spline implementation and smooth-continuity restriction. The universal approximation for arbitrary input dimensionality is proved and experimental verification is given, through which the mystery of ``black box'' of the solution space is largely revealed. The new proofs employed also enrich approximation theory.
format Preprint
id arxiv_https___arxiv_org_abs_2507_14177
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Understanding Two-Layer Neural Networks with Smooth Activation Functions
Huang, Changcun
Machine Learning
Artificial Intelligence
Numerical Analysis
68T07(Primary), 41A15(Secondary)
I.2.6; G.1.2
This paper aims to understand the training solution, which is obtained by the back-propagation algorithm, of two-layer neural networks whose hidden layer is composed of the units with smooth activation functions, including the usual sigmoid type most commonly used before the advent of ReLUs. The mechanism contains four main principles: construction of Taylor series expansions, strict partial order of knots, smooth-spline implementation and smooth-continuity restriction. The universal approximation for arbitrary input dimensionality is proved and experimental verification is given, through which the mystery of ``black box'' of the solution space is largely revealed. The new proofs employed also enrich approximation theory.
title Understanding Two-Layer Neural Networks with Smooth Activation Functions
topic Machine Learning
Artificial Intelligence
Numerical Analysis
68T07(Primary), 41A15(Secondary)
I.2.6; G.1.2
url https://arxiv.org/abs/2507.14177