Sequential Estimation of Gaussian Process-based Deep State-Space Models

We consider the problem of sequential estimation unknowns state-space and deep models that include functions latent processes models. The proposed approach relies on Gaussian are implemented via random feature-based processes. In these models, we have two sets unknowns, highly nonlinear (the values processes) conditionally linear constant parameters processes). present a method based particle filtering where integrated out in obtaining predictive density states do not need particles. also propose an ensemble version method, with each member having its own set features. With several experiments, show can track up to scale rotation.