Amphithéâtre Marguerite de Navarre, Site Marcelin Berthelot
Open to all
-

Abstract

For over thirty years, Simon Thorpe has been trying to understand how our brains manage to recognize images so quickly, despite the relative slowness of their neurons. He proposed that, given the speed of processing, much of it must be done with a unidirectional feed-forward processing wave. Since 2012, this type of architecture has been used by artificial systems with great success. But there are still other ideas from biology that can be a source of inspiration for AI. In this talk, he discusses two particularly interesting ideas absent in most artificial systems. Firstly, the fact that real neurons use " spikes " to encode information, and more specifically the order of neuron discharges. Then there's the fact that neurons can learn unsupervised repeating patterns.

Speaker(s)

Simon Thorpe

CNRS