Department of Mathematics

Thematic semester on Sequential Monte Carlo (SMC), Fall 2017

Plan for activities and a rough time plan
1. Aug 24–29, 2017, intensive PhD level course on SMC. Preliminary information is available at:
2. Aug 30 – Sept 1, 2017, SMC conference. Preliminary information is available at:
3. October, 2017, Two day focused workshop on the use of SMC in probabilistic programming.
4. September–December, 2017: Visitor program and seminar series.

Scientific description of the area
In recent years we have experienced an explosion of complex data-sets in a wide range of scientific fields, including machine learning, bioinformatics, epidemiology and climate science, to mention a few. In many of these applications, the mathematical models devised to accurately capture the dynamics and interactions of the data generating processes are complex and the only computationally feasible and accurate way to perform any kind of statistical inference is with Monte Carlo.
Sequential Monte Carlo (SMC) methods (also known as particle filters or particle methods) have over the past quarter of a century emerged as very successful tools for computational inference in complex statistical models. (For instance, a tutorial article (Arulampalam et al)  from 2002 is likely the most cited article in IEEE Transactions on Signal Processing, with more than 10,000 citations.) SMC was originally proposed as a tool for addressing the optimal filtering problem in signal processing and, indeed, in many potential applications areas it is still closely associated with optimal filtering. However, it has been realized that the methodology is in fact much more general and that it can be used to address the computational inference problems in a wide range of statistical models. Furthermore, over the past few years intensive research in this area has entirely breached the barriers between SMC and Markov Chain Monte Carlo, enabling a systematic integration of the two approaches that opens up for reaping the benefits from both. Together with the fact that SMC methods are naturally parallelizable, making this methodology well suited to modern computer architectures, this gives rise to an exciting period for advancements in the research area.

For further details about the thematic semester, please contact the organizers Thomas Schön and Fredrik Lindsten.