Combining Variational Methods with Sampling

The variational importance sampler is an hybrid algorithm which combines sampling with variational approaches to solve intractable integrals, particularly in the context of Bayesian inference.

Papers

Pre-prints of three papers which use or describe the approach are:

Reducing the Variability in cDNA Microarray Image Processing by Bayesian Inference by Neil D. Lawrence, Marta Milo, Mahesan Niranjan, Penny Rashbass and Stephan Soullier

Bayesian Processing of Microarray Images by Neil D. Lawrence, Marta Milo, Mahesan Niranjan, Penny Rashbass and Stephan Soullier

Variational Inference for Visual Tracking by Jaco Vermaak, Neil D. Lawrence and Patrick Pérez (PDF)

This paper gives an brief, non-exhaustive introduction to variational inference:

Variational Inference Guide by Neil D. Lawrence

Software for Microarray Analysis

We have developed software for image processing of cDNA microarrays using the variational importance sampler. The software is automated and provides a higher degree of consistency than other approaches. 

MPEG video of Software in action on a small portion of a slide. Purple lines indicate an initial rough grid layout. Blue ovals indicate where the algorithm is looking for the spot. Yellow ovals indicate where it finds it.

Release Information

Current release is 0.31.

As well as downloading the VIS software you need to obtain the toolboxes specified below.

Older Versions

Vs 0.3 Released 19 August 2003 - See readme.txt for changes. Vs 0.2 Released 24 June 2003 Vs 0.1 Released January 2003

Page updated on Mon Oct 29 00:18:02 2007