EC Bayes Seminar Series: Quan Vu (University of Wollongong)

The next EC* Bayes Seminar for 2022 presented by Quan Vu from the University of Wollongong.

Virtual attendees: A Zoom link will be emailed prior to event.

Location: QUT Gardens Point, room TBA

Title: Warped Gradient-Enhanced Gaussian Process Surrogate Models for Exponential Families with Intractable Normalizing Constants

Abstract: Bayesian inference with many exponential-family models, such as the Potts model, is often challenging because of the intractable normalizing constants that appear in the likelihood functions. Markov chain Monte Carlo (MCMC) methods to deal with these types of models, such as the exchange algorithm, require simulations at every iteration of the Markov chain, thus rendering inference computationally expensive. Surrogate models for the likelihood, often using Gaussian processes, have been developed to make inference computationally tractable. In this talk, we propose the use of a warped, gradient-enhanced Gaussian process surrogate model for the sufficient statistics, which jointly models the sample means and variances of the sufficient statistics and which uses warping functions to capture covariance nonstationarity in the input parameter space. We show that both the consideration of nonstationarity and the inclusion of gradient information can be leveraged to obtain a surrogate model that is better, in a mean-squared error sense, than the stationary Gaussian process, particularly in regions where the likelihood function exhibits a phase transition. We show that the surrogate model can be used to improve the effective sample size per unit time when embedded in exact inferential algorithms, such as importance sampling and delayed-acceptance MCMC.

To register please click: https://www.eventbrite.com/e/ec-bayes-seminar-series-quan-vu-university-of-wollongong-tickets-430745440207