I am an associate researcher in the Applied Games Group, working on methods for running inference algorithms on multiple processors. In particular I am targeting message passing algorithms such as Belief Propagation and Expectation Propagation. Many large scale applications of machine learning might benefit from parallelisation. One example application of this work is large scale models of genomic data.

I recently completed my PhD at David MacKay's Inference Group at the University of Cambridge. My thesis, 'Modelling Uncertainty in the Game of Go', presented a number of applications of machine learning to the game of Go. Go is an ancient Chinese game whose complexity has defeated attempts by Artificial Intelligence researchers to automate play. Typically in machine learning, uncertainty results from unpredictable aspects of the data which is often called 'noise'. In my work, I am primarily interested in uncertainty that results from a different source: limited computer speed (limited rationality). In Go, a board position in conjunction with the rules of the game contains all of the information necessary for perfect play. However, the sheer complexity of the game tree results in uncertainty about the future course of the game. I am interested in using probabilities (in the Bayesian sense) to represent and manage this uncertainty.