"Gibbs distributions and statistics of action potentials in neural networks"
Relatore: Bruno Cessac, Neuromathcomp, INRIA Sophia-Antipolis, France
29 Febbraio 2012 ore 15.00
|Modern advances in neurophysiology techniques such as micro-electrode arrays electro-physiology
have made it possible to observe simultaneously the spiking activity of assemblies of neurons.
The analysis of these recording ("spike trains") requires the development of accurate statistical methods.
We present such recent methods, based upon spatio-temporal constraints and maximal entropy principle, where spike statistics
is described by Gibbs distributions.
We show applications to spike trains recorded from ganglion cells of the salamander retina, in response to natural movies.
We discuss the problem of parameters fitting and parameters interpretation.
We also present recent mathematical results providing canonical forms of Gibbs distribution
in neural networks models. We argue that they could provide an alternative to the maximal entropy method
and related problem of parameters estimation and interpretation.