Starts 8 Mar 2017 11:00
Ends 8 Mar 2017 12:00
Central European Time
QLS Seminar: Hierarchical Bayesian modelling of volatile environments
ICTP
Central Area, 2nd floor, old SISSA building
Abstract:
While Bayesian inference on static quantities is widespread and has become standard in many contexts, Bayesian inference in volatile environments presents an additional set of challenges. These arise when inferring the continually changing states of an agent’s environment. To make accurate predictions about future observations, such an agent needs to model how states change. I will set out these challenges in detail and introduce recently developed methods for meeting them. In particular, I will go into the Hierarchical Gaussian Filter (HGF), a generic hierarchical model of inference on volatile and uncertain states. In this context, I will also give examples of neuroscientific studies where the HGF was used.
While Bayesian inference on static quantities is widespread and has become standard in many contexts, Bayesian inference in volatile environments presents an additional set of challenges. These arise when inferring the continually changing states of an agent’s environment. To make accurate predictions about future observations, such an agent needs to model how states change. I will set out these challenges in detail and introduce recently developed methods for meeting them. In particular, I will go into the Hierarchical Gaussian Filter (HGF), a generic hierarchical model of inference on volatile and uncertain states. In this context, I will also give examples of neuroscientific studies where the HGF was used.