Augmented Neural ODEs

About

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.

Emilien Dupont, Arnaud Doucet, Yee Whye Teh · 2019

Related benchmarks

Task	Dataset	Result
Image Classification	MNIST (test)	Accuracy98.2	894
Audio Classification	Speech Commands (test)	Accuracy91.1	65
Forecasting	MuJoCo 70% Dropped (test)	MSE0.057	27
Forecasting	MuJoCo 30% Dropped (test)	MSE0.056	27
Forecasting	MuJoCo Regular (test)	MSE0.055	27
Forecasting	MuJoCo 50% Dropped (test)	MSE0.057	27
Forecasting	Google stock data 70% Dropped scenario	MSE0.0023	11
Forecasting	Google stock data Regular scenario	MSE0.0023	11
Forecasting	Google stock data 30% Dropped scenario	MSE0.0025	11
Forecasting	Google stock data 50% Dropped scenario	MSE0.0029	11