Professor Steve R Gunn
Steve Gunn graduated from the University of Southampton in 1992 with a first class honours degree in Electronic Engineering. He obtained his PhD, from the University of Southampton in 1996. In 1996 he worked as a research fellow, in 1998 he became a lecturer, and in 2002 a senior lecturer in the Image, Speech and Intelligent Systems Research Group. In 2007 he was awarded a personal chair in the Information: Signals Images Systems Research Group at the University of Southampton.
His research covers two state-of-the-art areas, that of machine learning and computer vision. His work focuses on the development of techniques to convert these ill-posed inverse problems, to well-posed problems through the careful design of a function space with an appropriate prior.
His PhD work investigated the use of active contours (or 'snakes') to extract image boundaries. Active contour techniques use an energy minimisation framework to integrate prior generic constraints with the image data. The techniques differ significantly from conventional approaches in that they search for a local minimum. His work focussed on improving the robustness of active contours by considering the initialisation and parameterisation.
His recent research has involved the development of new machine learning algorithms based upon kernel methods and their application to a variety of domains. Kernel methods are rapidly replacing neural networks as the preferred tool for machine learning due to many attractive features: a strong basis from statistical learning theory; no computational penalty in moving from linear to non-linear models; the resulting optimisation problem is convex, guaranteeing a unique global solution and consequently producing systems with excellent generalisation performance, i.e. they learn extremely well.
His current research is focussed around the investigation of priors that induce a sparsity in a parameter space. This has advantages for learning in terms of speed, memory, representation and generalisation performance; these characteristics become increasingly important as larger data-sets are considered.
His interests include windsurfing, snowboarding, opera, reading and watching weather reports.