SOTAVerified

Latent ODEs for Irregularly-Sampled Time Series

2019-07-08Code Available1· sign in to hype

Yulia Rubanova, Ricky T. Q. Chen, David Duvenaud

Code Available — Be the first to reproduce this paper.

Reproduce

Code

Abstract

Time series with non-uniform intervals occur in many applications, and are difficult to model using standard recurrent neural networks (RNNs). We generalize RNNs to have continuous-time hidden dynamics defined by ordinary differential equations (ODEs), a model we call ODE-RNNs. Furthermore, we use ODE-RNNs to replace the recognition network of the recently-proposed Latent ODE model. Both ODE-RNNs and Latent ODEs can naturally handle arbitrary time gaps between observations, and can explicitly model the probability of observation times using Poisson processes. We show experimentally that these ODE-based models outperform their RNN-based counterparts on irregularly-sampled data.

Tasks

Benchmark Results

DatasetModelMetricClaimedVerifiedStatus
MuJoCoLatent ODE (ODE enc)MSE (10^-2, 50% missing)1.26Unverified
MuJoCoODE-RNNMSE (10^-2, 50% missing)26.46Unverified
PhysioNet Challenge 2012Latent ODE (ODE enc)mse (10^-3)2.23Unverified
PhysioNet Challenge 2012Latent ODE + Poissonmse (10^-3)2.21Unverified

Reproductions