Tuesday, 8 June 2004, 2pm:

<CENTER><B>Dynamical models of spike coding for the analysis of the visual
perception of motion</B>

<P>
Laurent Perrinet<BR>
CNRS, Universite P. Sabatier, Toulouse France

</CENTER>

Recent advances in the understanding of the mechanisms underlying the
visual perception of motion are rising from the convergence of
neuro-physiology, psycho-physics and modelling. I will describe a strategy
of sparse spike coding based on a simple implementation of an
over-complete network of Integrate-and-Fire neurons interconnected by
lateral connections. In particular, I will show how this algorithm leads
to a neural representation which is similar to the divisive normalization
computation.
This strategy allows to implement inferential mechanisms and I will
present an application of this model to the analysis of the visual
perception of movement. This model is implemented in close confrontation
with the psycho-physical data from the rapid dynamics of the visual
system.