Scope

Markov Chain Monte Carlo (MCMC) methods are of fundamental importance to various scientific fields, such as statistics, statistical physics, molecular dynamics and machine learning. The usefulness of these MCMC methods depends for a large part on their computational efficiency. Traditionally MCMC algorithms are in the particular class of reversible Markov chains. Intuitively this means that by watching a video of the algorithm as it progresses, it is impossible to tell whether the video is playing forward or backward in time.

In contrast, Non-reversible Markov chains can be thought of as having a certain sense of direction or vorticity. It turns out that these may have significant benefits in terms of computational efficiency over their reversible counterparts. It will be the aim of this workshop to explore:

• theoretical properties of non-reversible Markov chains that are important in determining their usefulness for MCMC;
• aspects of augmented state space algorithms such as Hamiltonian Monte Carlo and their relations to non-reversible Markov processes;
• theory and practice of design and implementation of efficient non-reversible Markov chains.

Keynote Speakers 

Organisers

• Joris Bierkens, CRiSM Research Fellow at the University of Warwick
• Gareth Roberts, Professor of Statistics at the University of Warwick

Correspondance should be sent to j.bierkens (at) warwick.ac.uk.

Link to:

Programme

Titles and abstracts

Registration

Poster Submission