From: KU Leuven; ESAT-STADIUS比利时鲁汶大学
?? How to model real-world multidimensional time series? especially, when these are sporadically observed data.
?? how to describe the evolution of the probability distribution of the data? ODE dynamics.
sporadically-observed time series: sampling is irregular both in time and across dimensions.
Evaluation on both synthetic data and real-world data.
Combine GRU-ODE and GRU-Bayes into GRU-ODE-Bayes model.
Introduction:
most methodology assumption: signals are measured systematically at fixed time intervals.
However, most real-world data is sporadic.
fixed time intervals data VS sporadic data.
How to model sporadic data becomes a challenge.
neural ordinary differential equation model; It opens the perspective of tackling the issue of irregular sampling.
interleave the ODE and the input processing steps; + GRU + Bayesian update network.
Performance metric: MSE, mean square error; NegLL, non-negative log-likelihood.
?? 可是他解决了一个什么问题还不知道，只知道 是model sporadical time series.