Stability-Informed Initialization of Neural Ordinary Differential Equations

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Westny, Theodor, Mohammadi, Arman, Jung, Daniel, Frisk, Erik
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910556609314816
author Westny, Theodor
Mohammadi, Arman
Jung, Daniel
Frisk, Erik
author_facet Westny, Theodor
Mohammadi, Arman
Jung, Daniel
Frisk, Erik
contents This paper addresses the training of Neural Ordinary Differential Equations (neural ODEs), and in particular explores the interplay between numerical integration techniques, stability regions, step size, and initialization techniques. It is shown how the choice of integration technique implicitly regularizes the learned model, and how the solver's corresponding stability region affects training and prediction performance. From this analysis, a stability-informed parameter initialization technique is introduced. The effectiveness of the initialization method is displayed across several learning benchmarks and industrial applications.
format Preprint
id arxiv_https___arxiv_org_abs_2311_15890
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Stability-Informed Initialization of Neural Ordinary Differential Equations
Westny, Theodor
Mohammadi, Arman
Jung, Daniel
Frisk, Erik
Machine Learning
Computer Vision and Pattern Recognition
This paper addresses the training of Neural Ordinary Differential Equations (neural ODEs), and in particular explores the interplay between numerical integration techniques, stability regions, step size, and initialization techniques. It is shown how the choice of integration technique implicitly regularizes the learned model, and how the solver's corresponding stability region affects training and prediction performance. From this analysis, a stability-informed parameter initialization technique is introduced. The effectiveness of the initialization method is displayed across several learning benchmarks and industrial applications.
title Stability-Informed Initialization of Neural Ordinary Differential Equations
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2311.15890