SMBp Group Meeting: Yuling Yao

Wednesday May 11, 2022, 2:00 PM → 3:00 PM America/New_York

3rd Floor Conference Room/3-Flatiron Institute (162 5th Avenue)

Speaker: Yuling Yao 

Topic: Multimodal sampling and free energy computation: A statistical perspective


Abstract: When working with multimodal Bayesian posterior distributions, Markov chain Monte Carlo (MCMC) algorithms have difficulty moving between modes, and default variational or mode-based approximate inferences will understate posterior uncertainty. And, even if the most important modes can be found, it is difficult to evaluate their relative weights in the posterior for it involves intractable some normalizing constant. 
I present two recent advances on multimodal sampling. The first method uses Bayesian stacking to combine multiple chains. The result from stacking efficiently samples from multimodal posterior distribution, minimizes cross validation prediction error, and represents the posterior uncertainty better than variational inference, but it is not necessarily equivalent, even asymptotically, to fully Bayesian inference.  The second method generalizes the classical simulated tempering idea to a continuous temperature. As a byproduct, we iteratively compute the normalizing constant as a continuous function of the temperature, which is itself important in Bayesian statistics as well as molecular dynamics.


The talk will be based on by recent papers 
Adaptive Path Sampling in Metastable Posterior Distributions (https://arxiv.org/abs/2009.00471), 
and Stacking for Non-mixing Bayesian Computations: The Curse and Blessing of Multimodal Posteriors (https://jmlr.org/papers/v23/20-1426.html)