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 | Rank | |
|---|---|---|---|---|
| Image Classification | MNIST (test) | Accuracy98.2 | 894 | |
| Forecasting | MuJoCo 70% Dropped (test) | MSE0.057 | 12 | |
| Forecasting | MuJoCo Regular (test) | MSE0.055 | 12 | |
| Forecasting | MuJoCo 30% Dropped (test) | MSE0.056 | 12 | |
| Forecasting | MuJoCo 50% Dropped (test) | MSE0.057 | 12 | |
| 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 | |
| Dynamics Prediction | Duffing oscillator noise level 0 | MSE (acceleration)3.15 | 7 |
Showing 10 of 17 rows