Sequential Monte Carlo methods for graphical models

Thomas B. Schön is Professor of the Chair of Automatic Control in the Department of Information Technology at Uppsala University. He received the PhD degree in Automatic Control in Feb. 2006, the MSc degree in Applied Physics and Electrical Engineering in Sep. 2001 and the BSc degree in Business Administration and Economics in Jan. 2001, all from Linköping University. He has held visiting positions with the University of Cambridge (UK), the University of Newcastle (Australia) and Universidad Técnica Federico Santa María (Valparaíso, Chile). He is a Senior member of the IEEE. He was awarded the Automatica best paper price in 2014 and the best PhD thesis award by The European Association for Signal Processing (EURASIP) in 2013. He received the best teacher award at the Institute of Technology, Linköping University in 2009. Schön's main research interest is nonlinear inference problems, especially within the context of dynamical systems, solved using probabilistic methods. He is active within the fields of machine learning, signal processing and automatic control. He pursues both basic research and applied research, where the latter is typically carried out in collaboration with industry.

 

Home page: user.it.uu.se/~thosc112/

 

 

In this talk I will introduce our new methods for inference in general probabilistic graphical models (PGMs). The key is a sequential decomposition of the PGM which provides a sequence of auxiliary distributions defined on a monotonically increasing sequence of probability spaces. By targeting these auxiliary distributions using Sequential Monte Carlo (SMC) methods we are able to approximate the full joint distribution defined by the PGM. I will also (briefly) introduce the underlying Sequential Monte Carlo methods (e.g. particle filters/smoothers), which are computational methods primarily used to deal with the state inference problem in nonlinear state space models. We are (since a few years back) seeing these methods finding new applications in more and more general model classes, for example PGMs. The systematic combination of SMC and MCMC, referred to as particle MCMC provides another powerful family of algorithms that we will touch upon. The first algorithms of this type were published in 2010 and since then we have (for very good reasons) witnessed a rapidly growing interest in these algorithms.