Fig. 1: Main ideas.
From: State estimation of a physical system with unknown governing equations
a, Given noisy time-series data and an observation model, our goal is to estimate the underlying state and any missing terms in the forward model. b, We specify a hierarchical prior over the state and the governing equations and then infer an approximate posterior over the state and forward model using SVI. c, We arrive with a state estimate, a method for generating forecasts with uncertainty estimates and a Bayesian estimate for missing terms in the governing equations.
