Workshop on Recent Advances in Bayesian Computation
(20 - 22 September 2010)

Organizing Committee · Visitors and Participants · Overview · Activities · Venue

 

Organizing Committee

 

Chair

• David Nott (National University of Singapore)

 

Members

• Leontine Alkema (National University of Singapore)
• Alex Cook (National University of Singapore)
• Robert Kohn (University of New South Wales)

 

Visitors and Participants

 

• Overseas visitors
• Local visitors
• Graduate students
• Registered local participants

 

Overview

 

Over the last 15 years there has been an explosion in the use of Bayesian methods in applied statistics. Due to advances in technology for data collection in many fields of science, engineering and the social sciences, applied statisticians increasingly have to deal with problems of combining data and information from different sources. This leads naturally to the use of richly structured hierarchical models, and advances in Bayesian computational methods have meant that a Bayesian approach is often the most easily implemented one for inference in such models. The purpose of this workshop is to bring together leading researchers in the area of Bayesian computational methods to discuss challenges and opportunities in the area, with a focus on dealing with large data sets.

 

The breakthrough technology that has allowed routine use of Bayesian methods in complex problems is Markov chain Monte Carlo (MCMC). Although MCMC has a long history of use in statistical physics its potential for general purpose Bayesian inference in statistics was only fully realized in the 1990's, starting with Gelfand and Smith (1990) and following some earlier applications in spatial statistics (Geman and Geman, 1984). New theoretical developments in MCMC are now appearing more slowly than previously but there is continuing widespread use of MCMC in applications. Although MCMC has allowed statisticians to handle complex models often with hundreds or thousands of parameters, one weakness is that MCMC is very computationally intensive, and for large data sets the use of MCMC methods may be completely infeasible. In short, while MCMC allows computation for complex models with small to moderate sized data sets, computation for complex models and very large data sets remains a challenge.

 

Many alternative strategies for computation with large datasets and complex models are being actively explored. These include clever methods for sampling large datasets, adaptive MCMC methods that attempt to improve on traditional MCMC approaches, advanced importance sampling techniques and deterministic methods such as variational approximation (developed actively within the machine learning community) and Laplace approximation. It is likely that advances in Bayesian computation for large data sets will involve at least to some extent imaginative combinations of the approaches discussed above. A key goal of the workshop is to look at the field of Bayesian computation very broadly and to bring together researchers from different areas to enable a cross-fertilization of ideas.

 

Gelfand, A. and Smith, A. (1990). Sampling based approaches to calculating marginal densities. J. American Statist. Assoc., 85, 398-409.

 

Geman, S. and Geman, D. (1984). Stochastic relaxation, Gibbs distributions and the Bayesian restoration of images. IEEE Trans. Pattern Anal. Mach. Intell., 6, 721-741.

 

Activities

 

The workshop will be held over 3 days from 20-22 September 2010. There will be talks from approximately 10 overseas invited speakers and several local speakers. Although all the talks are by invitation, there will also be a poster session on the afternoon of 21 September which is open to all attendees.

Monday, 20 Sep 2010
09:00am - 09:20am	Registration
09:20am - 09:30am	Opening Remarks Louis Chen, Institute for Mathematical Sciences David Nott, National University of Singapore
	Chair: Robert Kohn, University of New South Wales
09:30am - 10:15am	The r-inla.org project: an overview HÃ¥vard Rue, Norwegian University of Science and Technology, Norway
10:15am - 10:45am	--- Coffee Break ---
10:45am - 11:30am	EM, variational bayes and expectation propagation Mike Titterington, University of Glasgow, UK
11:30am - 01:30pm	--- Lunch ---
	Chair: Chenlei Leng, National University of Singapore
01:30pm - 02:15pm	Skew-normal variational approximations for Bayesian inference John Ormerod, University of Sydney, Australia
02:15pm - 03:00pm	Variational bayes for elaborate distributions Matt Wand, University of Wollongong, Australia
03:00pm - 03:30pm	--- Coffee Break ---
	Chair: Minh Ngoc Tran, National University of Singapore
03:30pm - 04:15pm	Large-scale Bayesian logistic regression David Madigan, Columbia University, USA

Tuesday, 21 Sep 2010
09:15am - 09:30am	Registration
	Chair: Leontine Alkema, National University of Singapore
09:30am - 10:15am	Bayesian hypothesis testing in latent variable models Jun Yu, Singapore Management University
10:15am - 10:45am	--- Coffee Break ---
10:45am - 11:30am	Optimizing MCMC algorithms in high dimensions: a new perspective. Natesh Pillai, Harvard University, USA
11:30am - 01:30pm	--- Lunch ---
	Chair: Yanan Fan, University of New South Wales, Australia
01:30pm - 02:15pm	Bayesian computation on graphics cards Chris Holmes, University of Oxford, UK
02:15pm - 03:00pm	Finite dimensional simulation methods for infinite dimensional posteriors Jim Griffin, University of Kent, UK
03:00pm - 03:30pm	--- Coffee Break ---
	Chair: Scott Sisson, University of New South Wales, Australia
03:30pm - 04:15pm	Variational Bayes for spatial data analysis Clare McGrory, Queensland University of Technology, Australia
04:15pm - 05:00pm	Regression density estimation with variational methods and stochastic approximation David Nott, National University of Singapore

Wednesday, 22 Sep 2010
09:15am - 09:30am	Registration
	Chair: Alex Cook, National University of Singapore
09:30am - 10:15am	Recent advances in ABC (Approximate Bayesian Computation) methodology Jean-Michel Marin, Université Montpellier 2, France
10:15am - 10:45am	--- Coffee Break ---
10:45am - 11:30am	Adaptive Monte Carlo sampling and model uncertainty Merlise Clyde, Duke University, USA
11:30am - 01:30pm	--- Lunch ---
	Chair: Siew Li Linda Tan, National University of Singapore
01:30pm - 02:15pm	The expected auxiliary variable method for Bayesian computation Arnaud Doucet, University of British Columbia, Canada
02:15pm - 03:00pm	Flexible modeling of conditional distributions using smooth mixtures of asymmetric student t densities Robert Kohn, University of New South Wales, Australia
03:00pm - 03:30pm	--- Coffee Break ---
	Chair: Roman Carrasco, National University of Singapore
03:30pm - 04:15pm	Help! Fitting process-based models to infectious disease data, Bayesianly Alex Cook, National University of Singapore
04:15pm - 05:00pm	Adaptive optimal scaling of Metropolis-Hastings algorithms Scott Sisson, University of New South Wales, Australia

Students and researchers who are interested in attending these activities and who do not require financial aid are requested to complete the online registration form.

The following do not need to register:

• Those invited to participate.
• Those applying for financial support.

Venue

 

• IMS Auditorium
• Detailed instruction on how to get to the institute by public or private transport

 

Organizing Committee · Visitors and Participants · Overview · Activities · Venue