skip to content

Fast Bayesian Boolean Matrix Factorisation

Presented by: 
Chris Holmes
Wednesday 5th July 2017 - 15:30 to 16:15
INI Seminar Room 1
Boolean matrix factorisation decomposes a binary data matrix into an approximating Boolean product of two low rank, binary matrices: one containing meaningful patterns (signatures), the other quantifying how the observations can be expressed as a logical combination of these patterns.  

We introduce a probabilistic model for Boolean matrix factorisation, termed the “OrMachine”, and derive a Metropolised Gibbs sampler that facilitates efficient parallel posterior inference on commodity hardware. On real world and simulated data, our Bayesian method provides state of the art performance for Boolean matrix factorisation and matrix completion. The method supports full posterior inference, which is important in applications, for example in controlling false positive rates in collaborative filtering and, crucially, improves the interpretability of the inferred patterns. The proposed model and computation scale to large datasets as motivated by an analysis of single cell gene expression data recording measurements from 1.3 million mouse brain cells across 11 thousand genes. 
The video for this talk should appear here if JavaScript is enabled.
If it doesn't, something may have gone wrong with our embedded player.
We'll get it fixed as soon as possible.
University of Cambridge Research Councils UK
    Clay Mathematics Institute London Mathematical Society NM Rothschild and Sons