Computer Science graduate wins national award for dissertation

A Manchester graduate has been awarded the British Computer Society Distinguished Dissertation Award 2013 for the best PhD dissertation in Computer Science.

Dr Adam Pocock’s thesis, ‘Feature Selection via Joint Likelihood,’ focused on the problem of identifying relevant and irrelevant variables in statistical machine learning* problems. He was supervised by Drs Gavin Brown and Mikel Lujan.

Adam is now a researcher at Oracle Labs in the Information Retrieval and Machine Learning group in Burlington, Massachusetts.

Adam said "I'm really surprised and pleased to have won this award; it’s an amazing end to my eight years spent at Manchester, through a BSc, MSc and a PhD. I wouldn't have managed without the help and support from my friends in the lab, the staff in CS and especially my supervisors Gavin and Mikel."

Adam’s success follows on from that of Computer Science PhD postgraduate Matthew Horridge, who won the CPHC/BCS Distinguished Dissertation Competition in 2012, resulting in success for the School of Computer Science two years running. Since the award’s inception in 1990 Manchester Computer Science postgraduates have received the award four times.

Adam’s award is the third prestigious national award for the School of Computer Science this semester – at the 2013 Science, Engineering and Technology awards in September his supervisor Gavin Brown received the SET Lecturer of the Year award and another of Gavin’s students’, Laura Howarth-Kirke, received overall Science, Engineering and Technology Student of the Year, for her project entitled, ‘Learning and Recognising Human Gestures using the Microsoft Kinect’.

Ends

Notes for editors

*What is Machine Learning and Feature Selection?

In solving any given problem, some pieces of information are relevant, some are irrelevant, and some only show their worth in the context of others. For example, in predicting the price of a car: the make/model are relevant, the colour is probably irrelevant, and the age of the vehicle is probably relevant only in the context of the number of miles the car has been driven. You know this because you know something about cars. But what about harder problems: like predicting whether someone will have a relapse of a particular cancer, or knowing what factors influence whether a customer will purchase a given product? In Computer Science, the field of Machine Learning refers to this as "feature selection". There have been many solutions proposed to this generic problem, in particular many in an area called "information theoretic methods". Dr Pocock's PhD thesis (and other work with Dr Brown and Dr Lujan) provided a unifying mathematical framework which explained the success of over 20 years of academic publications in this area, meaning each method over the past two decades can be understood from one single unifying principle.

For further information contact:

Aeron Haworth
Media Relations
Faculty of Engineering and Physical Sciences
The University of Manchester

Tel: 0161 275 8387
Email: aeron.haworth@manchester.ac.uk