University of Tartu - ©2011 Rafik Chaabouni - Last update: 08.06.2013 18:08
Date: 20/06/2013 Location: J. Liivi 2, room 317 (next to the coffee room)
Speaker: Vincent Gripon (Invited speaker from Télécom Bretagne)
Title: Calculating using associative memories
Abstract:
Associative memories are datastructures that can store then retrieve
elements given a part of their content. They are used in many application
fields at the hardware level, but are less frequent at the software level.
As a matter of fact, celebrated models such as CAMs in electronics and
Hopfield neural networks in neuroscience are not scalable. We introduce a
new model of associative memories that achieve memory efficiency at a
constant factor from the optimal. They offer a considerable gain in terms of
diversity (number of stored messages) and capacity (amount of information
stored) compared to state-of-the-art techniques. Therefore, they appear to
be interesting perspectives in the fields of artificial intelligence and
machine learning.
Bio:
Vincent Gripon obtained a PhD in computer science at Télécom Bretagne in
2011. He is mainly interested in bio-inspired algorithms, datastructures,
and iterative algorithms. He is also the co-creator and co-organizer of an
online programming contest named TaupIC.