Abstract
State-space models are widely used for the analysis of time-series data but full Bayesian inference is still elusive. I will review some applications of state-space models and recent endeavours towards efficient Monte Carlo sampling, in particular using Particle filtering and more recent Particle Markov Chain Monte Carlo methods. I will discuss the theoretical scalability of these methods with respect to the length of the observed time-series. Our theoretical results on their efficiency align well with many documented instances of their effectiveness based on extensive numerical studies. The talk will conclude with some open challenges to be pursued.
Brief Biography
Sumeetpal S. Singh received his Ph.D. from the University of Melbourne in Dec. 2002 and is currently a University Reader and the Head of the Signal Processing and Communications group at the Department of Engineering in Cambridge University. He is also a Fellow of Churchill College and an Academic Fellow of the Alan Turing Institute (since 2016.) His research interests include Multi-object tracking, Bayesian analysis of state-space models, Reinforcement Learning and more generally Computational Statistics. He was the recipient of the IEEE Barry Carton award in 2010 and is currently an Associate Editor of Statistics and Computing (Springer), a founding Executive Editor of a new journal on Data Centric Engineering by Cambridge University Press and an Associate Editor of Foundations of Data Science (AIMS).