Gaussian process inference for stochastic continuous time dynamical systems

Lecture
Datum: 
12. September 2014
Beginn: 
10:15
Raum: 
CITEC 2.015

Abstract

We study the learning of continuous time dynamical systems described by nonlinear differential equations. To explore the phase space of the dynamics, additional white noise is added.
Using Gaussian process prior distributions over the nonlinear forces, we develop an approximate Bayesian inference method which allows us to estimate the unknown forces from discrete time observations of trajectories.