A Bayesian approach to multicanonical Markov chain Monte Carlo

Add to your list(s) Download to your calendar using vCal

• Dr Jes Frellsen, CBL, CUED
• Thursday 29 January 2015, 14:00-15:00
• Board Room, CUED.

If you have a question about this talk, please contact Fredrik Lindsten.

The Markov chain Monte Carlo method is one of the most important tools for approximate inference. However, the canonical ensemble, known as the posterior distribution in Bayesian inference and the Boltzmann distribution in physics, is in many cases difficult to sample from due to slow convergence and poor mixing of the Markov chain. The so-called multicanonical ensemble can potentially alleviate this problem by constructing a new distribution that performs a random walk between different values of the energy function. Estimates for posterior averages and partition functions can then be obtained by reweighting. The main challenge in applying the multicanonical ensemble is that one has to solve an inference problem on the density of energy under the prior. In this talk I will present a Bayesian approach to solve this problem using a Gaussian Process prior. I will present preliminary results on spin models and discuss applications in protein folding.

This talk is part of the Signal Processing and Communications Lab Seminars series.

This talk is included in these lists:

• All Talks (aka the CURE list)
• Board Room, CUED
• Cambridge Centre for Data-Driven Discovery (C2D3)
• Cambridge University Engineering Department Talks
• Cambridge talks
• Centre for Smart Infrastructure & Construction
• Chris Davis' list
• Computational Continuum Mechanics Group Seminars
• Featured lists
• Information Engineering Division seminar list
• Interested Talks
• School of Technology
• Signal Processing and Communications Lab Seminars
• Trust & Technology Initiative - interesting events
• bld31
• ndk22's list
• ob366-ai4er
• rp587

Note that ex-directory lists are not shown.