Augmented Neural ODEs

Emilien Dupont, Arnaud Doucet, Yee Whye Teh
4/2/2019
stat.MLcs.LG

Official code: EmilienDupont/augmented-neural-odes (560 stars, MIT, last commit 2023-02-09).

Abstract

We show that Neural Ordinary Differential Equations (ODEs) learn representations that preserve the topology of the input space and prove that this implies the existence of functions Neural ODEs cannot represent. To address these limitations, we introduce Augmented Neural ODEs which, in addition to being more expressive models, are empirically more stable, generalize better and have a lower computational cost than Neural ODEs.

View on arXivView PDF

Code Implementations(1)

EmilienDupont/augmented-neural-odesOfficial100%
MIT

Pytorch implementation of Augmented Neural ODEs :sunflower:

56090Python, Jupyter NotebookMay 19, 20193 years agoMIT

Similar Papers

Function Spaces Without Kernels: Learning Compact Hilbert Space Representations

77% match2025

Improving Memory Efficiency for Training KANs via Meta Learning

77% match1 repo72025

Latent Functional Maps

77% match2024

Learning Fair Representations for Kernel Models

76% match2019

Non-Parametric Representation Learning with Kernels

76% match2023

When Representations Align: Universality in Representation Learning Dynamics

76% match2024

Cite this paper

@article{dupont2019augmented,
  title  = {Augmented Neural ODEs},
  author = {Emilien Dupont and Arnaud Doucet and Yee Whye Teh},
  year   = {2019},
  eprint = {1904.01681},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1904.01681}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1904.01681.svg)](https://codeofpaper.com/paper/1904.01681)

Discussion