Functional networks from inverse modeling of neural population activity

The availability of large-scale neural multi-electrode or optical recordings make now possible the modelling of the simultaneous activities of tens to thousand of neurons. One promising approach relies on the inference of detailed functional connectivity between the recorded cells, that is, of an effective coupling network reproducing the correlation structure of the spiking events. Here we report some recent applications of those approaches to retinal, hippocampal, and cortical data, illustrating in particular how functional coupling networks may be useful to decode complex brain representations, and how their changes may be tracked in behaving animals, with a possible connection to behavioral learning. Statistical, theoretical, and neurobiological issues raised by the inverse modeling of population activity are discussed.