Neural Particle Filter and Beyond

Simone Surace, University of Zürich, ETH Zürich and University of Bern, Switzerland 2.9.0.1410:15 - 11:00

Perception can be seen as unconscious inference in a dynamically changing environment and formalized as nonlinear Bayesian filtering. A heuristic search for a scalable and biologically plausible algorithm led us to an interacting particle filter that is very similar to existing algorithms in data assimilation. Taking inspiration from one of these algorithms, the feedback particle filter, which is obtained from a control theory framework, we explore new ideas for optimality principles for particle filters.

Invited by Sebastian Reich