ELIZABETH@OZ.AI.MIT.EDU.UUCP (11/09/87)
From: Peter de Jong <DEJONG%OZ.AI.MIT.EDU@XX.LCS.MIT.EDU>
Subject: Cognitive Science Calendar [Extract - Ed]
[Forwarded from the IRList digest.]
DYNAMICAL CONNECTIONISM
Elie Bienenstock
Universite de Paris-Sud
Wednesday, 11 November
E25-406, 12:00
In connectionist models, computation is usually carried out in a space
of activity levels, the connectivity state being frozen. in contrast,
dynamical connectionist models manipulate connectivity states. For
instance, they can solve various graph matching problems. They also
have the typical associative memory and error-correcting properties of
usual connectionist models. Applications include invariant pattern
recognition; dynamical connectionist models are able to generalize
over transformation groups rather than just Hamming distance. It is
proposed that these principles underlie much of brain function; fast-
modifying synapses and high-resolution temporal correlations may
embody the dynamical links used in this new connectionist approach.