Statistics Seminars: Hybrid Monte Carlo in High Dimensions
1 November 2010 15:15 in CM221
The Hybrid Monte-Carlo algorithm employs Hamiltonian dynamics to provide non-local, non-symmetric proposals for MCMC algorithms. I will briefly describe the algorithm and some of its properties before I present our investigations on the behaviour of the algorithm in high dimensions. Additionally, I will present a modification of the standard method which gives rise to a Hybrid Monte-Carlo algorithm on a Hilbert space and discuss its practical implications.
Host: Umberto Picchini
Contact email@example.com for more information