Overview
I am mainly interested in applications of Stastical Machine learning techniques to practical problems. The applications I have worked on can be divided into 3 main areas:
Brain Computer interfaces (BCI)
A BCI is designed to allow a person to control a machine by simply thinking. Machine learning is critically important for practical BCIs as it allows allows the machine to rapidly adapt to the unique signals generated by each individual user. I am currently investiagating how to train subject-specfic feature extraction algorithms, and how to adapt the BCI on-line to changes in the subjects brain state.
Real-Time decision making and Bounded Rationality.
Computational agents in the "real-world" must be able to make quick, complex, goal-directed, real-time decisions in the context of limited resources and uncertainty about the environment. Towards this end my PhD thesis work investigated how learning and decision theoretic control of reasoning (meta-control) could be used to learn useful information about the environment and direct computation effort where it was most useful.
Generic Image Categorisation.
This is the problem of recognising that an image contains particular types of objects, such as chairs, trees, cars, phones etc. As for BCI machine learning is critical to this problem to adapt to the huge varability in images within the same class, caused by for example, changes in: object instance, orientation, lighting, background.. To address this problem we combined SIFT interest point detectors with specialised kernels for working with sets of objects.