Towards an Interface for Streaming Stan

We are developing an extension of Stan that makes it possible to use Stan in the context of streaming data. We use the Stan syntax to define a dynamic model and likelihood for each time-step. We then pass a stream of measurements one at a time to the extended version of Stan, which then runs a particle filter (which we hope to optimise in the future) to produce a set of weighted particles representing the uncertainty. The aim of this talk is to provide potential users with sight of how models are defined and the code is used. The intent is that feedback will inform our ongoing development of what we hope will be an augmentation of Stan that the Stan community will find useful in contexts where Stan cannot currently be applied.

Presenter biography:
Simon Maskell

Simon Maskell is Professor of Autonomous Systems at the University of Liverpool and leads a number of projects looking to extend and apply the state-of-the-art in numerical Bayesian inference.