Research interests
My interests are centered around knowledge. How is knowledge acquired and how is it used to form expectations that guide our behaviour and that influence further acquisition of knowledge about the world. I am very much interested in the brain, ie in neural solutions to such problems. For this reason I am doing work at cellular level, trying to further our understanding of single-cell computation and at the level of populations of neurones. Most recently, at a more abstract level, I have started working on models of psychiatric diseases, and this is what I mainly concentrate on now.
Computational psychiatry
Our approach will be based around the fact that most psychiatric disorders involve neuromodulators and that recent work has yielded considerable insight into the normative function of these same neuromodulators. Thus, it is now possible to approach the dysfunctions in psychiatric disorders from a normative understanding of normal function, hence the term "computational psychiatry".
Population coding
This work looks at population coding from an angle which has been surprisingly neglected: time. What happens to the structure of the code if stimuli are not static, but move around? We show that interpretation of spikes in a dynamic, fast-changing world is impossible without a prior over how stimuli change. Depending on the nature of this prior, the code can be computationally very unwieldy, but it can still be handled approximately by recurrent neural networks.
The work on population coding was done with Peter Dayan, and with Rich Zemel and his student Rama Natarajan at the University of Toronto. Recently, I've been exploring the possibility of testing some predictions psychophysically with Marc Ernst. Initially, I also worked with Sophie Deneve on line attractors.
Single-cell models
Building detailed, biophysically realistic single-cell models remains a major challenge. The nature of the parametrization of these models results in extremely complex, non-linear interactions between the various parameters. We here build on recent advances in imaging techniques, which will soon allow access to the transmembrane voltage at many points throughout a cell's dendritic arbor. Given this rich information, and some more constraints, it is possible to set many of the parameters of interests in an automatic way.
The work on single-cell models is in collaboration with Liam Paninski, who is now at Columbia and Misha Ahrens who's also a PhD student here.
Medicine
Acoustic neuromas are slowly growing tumours of the auditory nerve (VIII). They are clinically very hard to detect and it may be dangerous to miss them. The "Gold Standard" for diagnosis at present is MRI scanning, which is prohibitively expensive. In an attempt to provide an alternative, we used acoustic data from patients that had undergone MRI scanning to predict the outcome of the scan.
This work has mostly been in collaboration with Reza Nouraei.
Science and Art
I am very interested in how artistic and scientific understanding relate. I am currently exploring such issues with Rachel Warrington and Cheryl Frances-Hoad.