Download presentation
Presentation is loading. Please wait.
1
Variational Methods TCD Interests Simon Wilson
2
Background We are new to this area of research – so we can’t say very much about it – but we’re enthusiastic! Our interest is in the use of variational methods for implementing Bayesian inference; Current literature in statistics concentrates on the “variational Bayes” approach to approximating posterior distributions; This is an iterative procedure: –It tends to be faster than MCMC –It tends to underestimate the posterior variance
3
Variational Bayes Approximation of a posterior distribution p( | data) by some other (more amenable) function q( ); Objective function - KL distance “Mean field approximation” usually assumed: iterative updating of each q i developed
4
Research Questions How to exploit other aspects of variational methods to propose new approximative techniques for Bayesian inference; Can we exploit parallel computing for high- dimensional problems? –Variational Bayes not necessarily good for this (it’s iterative) Motivate through image applications; Funding proposals currently in progress – hope to have Ph.D. student from October 06 (will know by May).
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.