Jimmy Olsson: Statistical inference and learning in general state-space models

Abstract

As one of the most powerful modeling tools of modern statistics, dynamical state-space models—also known as general state-space hidden Markov models—are today widely used in diverse areas such as pattern recognition, signal processing, bioinformatics, finance, and many more. In this talk, I will give a brief overview of statistical inference and learning in this kind of generative models, including statistical theory and computational algorithms. A special focus will be on sequential Monte Carlo methods, which today constitute a standard tool for sampling online from sequences of distributions, including those state posteriors that are of vital importance for inference and learning in state-space models. In this context, I will describe some of my own contributions to this field and present some open problems.