Lab Lunch: 11 June 2019 - Jane Hillston

Title: Statistical Abstraction for Multi-scale Spatio-temporal Systems

Abstract:

Spatio-temporal systems often exhibit multi-scale behaviours, in which response to a physical environment drives internal processes which in turn influence spatial behaviour.  However such systems present formidable challenges for computational modelling and analysis.  I will present a prototypic scenario in which spatially distributed agents decide their movement based on external inputs and a fast-equilibrating internal computation: bacterial chemotaxis.  Illustrated through this example, I will propose a generally applicable strategy of model abstraction based on statistically abstracting the internal system using Gaussian Processes, a powerful class of non-parametric regression techniques from Bayesian Machine Learning.

Joint work with Michalis Michaelides and Guido Sanguinetti.