Current Research Projects
My main areas of research are statistical machine learning and pattern recognition. Specific lines of current research are summarised below.
Causal discovery in disease data
People: Ernest Mwebaze, John Quinn, Michael Biehl (University of Groningen, Netherlands).
It is sometimes thought to be impossible to discover causes of events without any background knowledge or the ability to do experiments. However, the field of inferring causes and effects with purely observational data is developing. Correlation does not directly imply causation, but some patterns of association make particular causal relationships more likely than others.
This work is focused on developing fast methods to find strong causes and effects related to a target variable from a large set of covariates. This is useful (1) for gaining insight into a domain, and (2) for prediction of the effects of interventions. We are particularly interested in applying this to data collected in Uganda concerning prevalence of disease and the outbreak of epidemics such as cholera and ebola. This analysis could confirm or disconfirm our ideas about climatic, demographic and environmental factors which are thought to influence such events. An indication of the relative strengths of different causes can also help in predicting the efficacy of different eradication policies.
New (04/12/08): Entry to NIPS 2008 causal discovery competition received honourable mention for "significant advance on the REGED dataset".
Novelty detection in sequences
People: Martin Mubangizi, John Quinn, Chris Williams (University of Edinburgh, UK), Neil McIntosh (University of Edinburgh, UK).
There are some times when it is useful to know when things are “not normal”, for example when monitoring a patient in intensive care, flying a plane or measuring disease patterns in regions of a country. To work out what is abnormal first involves coming up with a model for what is normal. Inference of novelty could then be described as choosing whether to believe the model or to believe the data, which we can do using Bayesian methods.
We have developed methods which are effective in finding clinically significant unusual sequences in physiological monitoring data (more information, publications and media coverage here). We are now working on applying these types of methods to disease data in Uganda, as above. Novelty detection in a biosurveillance setting is desirable in order to give early warning of uncharacteristic changes which might lead to an epidemic. Work is also continuing on monitoring in neonatal intensive care, with the aim of making baby alarms more specific.
New (19/12/08): work supported by an IBM Faculty Award, for the project "Machine learning techniques for prediction of cholera outbreaks".
Traffic monitoring and vehicle tracking
People: Rose Nakibuule, John Quinn, Florence Tushabe, Michael Wilkinson (University of Groningen, Netherlands).
Any machine learning visitors coming to Makerere (encouraged) will be struck by the green hills and congenial climate, but also by the rampant, chaotic traffic in Kampala. This state of affairs has motivated our work in robust traffic monitoring and vehicle tracking. We are initially using static images, then moving to video streams from CCTV.
An idea currently being explored is to measure of the level of congestion on a road using morphological analysis, i.e. the distribution of different types of shapes in the scene. The local environment presents different challenges to those normally considered in traffic monitoring work. Compared to approaches designed for constrained motorway environments, we have to be able to deal with all manner of traffic types, including vehicles travelling the wrong way or amongst pedestrians, and in cluttered scenes including people and animals.