I am concerned with modelling the following brain structures: Prefrontal Cortex (PFC), Hippocampus (HS) and Basal Banglia (BG). The global framework is a bottom-up approach. We first define some reflex behaviors before learning more complex behaviors.
Our models are based on the building of Place Cells (PCs). PCs are observed in rodents and are cells that fire in a particular location in the environement.
Of particular interest is that we hypothetize the existence of "transition cells" in HS (CA3 and CA1 regions).
In parallel with the creation of PCs and transition cells in HS, a cognitive map is built in PFC. This map is used for planning movements to a goal location by biasing the proposed movements from HS.
For the time being, PCs and TCs rely on visual information. It is currently under investigation whether these cells may also be created by multisensory signals (touch, sound), or even "context".
These projects are supported by ANR project NEUROBOT.
I have been working with C. Lavandier (MRTE, UCP) for 4 years. We work on the classification of sound perception environment (both metric and temporal) with Kohonen maps, and the prediction of the quality of sound ambiance with multi-layer perceptrons.
This work was supported by the QASOAR contract, and is currently funded by the CartAsur project.
In parallel with the neurobiological models, I also study Neural Networks in a dynamical system perspective. In particular, I study Random Recurrent Neural Networks (RRNN) both with rate coding neurons and spiking neurons.
Of particular interest is the chaotic dynamics exhibited by both networks. This enables to store spatio-temporal inputs through hebbian learning in RRNN for instance. During a sabbatical at the Brain Lab at UNR (Reno, NV, USA). I studied RAIN (Random Asynchronous Irregular firing Networks).
These sudies were supported by ANR project ASTICO, PEPS ST2I MARTINE, DARPA project Synapse.
