Aller au contenuAller au menuAller à la recherche

Hervé FREZZA - BUETBrain-inspired system as a challenge for computer science

Abstract :


Today, it is clear that neuro-sciences can rely on high performance computer simulations in order to handle complex and more realistic models of neural systems.
What is sometimes less clear in the scientific community is that computing is much more than a powerful tool for scientists, since computation is indeed the central object of its own dedicated science. Moreover, even in the field of computer science itself, it is not always known that understanding the brain can bring the new paradigms we need for enlarging the set of what we can do with computers.
The goal of the talk is to present to biologists and neuro-scientists some current limitations of the "science of computation", thus isolating the exchange area from which biology can percolate into computer science. Those limitations are mainly situated systems like autonomous and adaptive robots, since this application field is currently an open problem raised in the 50's.
Main ideas of the paper given in reference will be presented during the talk as an example of what biology can bring to computer science.
It proposes a computational architecture that implements mainly cortically inspired paradigms, as joint multi-modal self-organization, but also first steps in handling reinforcement learning in such a context. This latter point is crucial in our field for moving further toward cognition, but difficult to integrate in our current neural systems (this is current work). This concerns relations between procedural learning and selection of action as well as the management of reward through time. Such topics resonate with basal ganglia studies, for which a feed-back from the audience is of course expected.

Selected publications

Olivier Ménard and Hervé Frezza-Buet. Model of multi-modal cortical processing: Coherent learning in self-organizing modules. Neural Networks, 18(5-6):646-655, 2005.

Thomas Boraud