Urs Köster

Home menu

Research interestes and activities

Just like almost everybody in the field, I am originally coming from a physics background, but my interest in neuroscience motivated me to switch to a PhD in Computer Science. I am interested in how the brain works, and I hope to achieve some insight into that by studying computational models that simulate the behavior of neural circuits in Visual cortex. As all of cortex has a fairly similar structure, my hope is that once the rather accessible field of vision is understood, this will generalise to other areas.

My research so far has concentrated on statistical models of early visual processing, owing to my supervisor Aapo Hyvarinen, who spend a lot of his time using ICA and related methods to model the behavior of cells in primary visual cortex. My contribution to this so far has been in providing additional evidence that the principle of statistical independence on natural images can not only explain simple cell receptive fields but is also sufficient to produce complex cells.

I am very interested in modeling of higher levels of visual processing, for example cue invariant boundary perception and possible representations of these higher order features. Simply including more layers of features in our models mainly presents computational challenges, but there certainly is more to it. So I am trying to figure out ways how distributed information coding in population of neurons can be included into these models. The dynamics of recurrent network models without stable states - tempting stuff for the physicist that I am - can certainly yield interesting aspects to extend the feed-forward models that we have been working on in the past.

My CV can be found here.