Abstract

Reactive synchronous languages are now a standard industry tool for critical embedded systems. Designers write high-level specifications by composing streams of values. Such systems typically evolve in noisy environments that can only be observed through noisy sensors. In this talk, I will present ProbZelus, a synchronous language extended with probabilistic constructs for Bayesian reasoning to model uncertainty. ProbZelus programs describe state-space models interacting with an observable environment. At runtime, an inference engine estimates the parameters of the model from observations to produce a stream of distributions. I will then detail the semi-symbolic inference algorithms that we use for efficient streaming inference which combine approximate sampling methods and exact symbolic computations.