Lecture by James McClelland

Complementary Learning Systems: How brain systems work together to support memory and learning (Maastricht)

25 september 2014 van 13:00 tot 14:30 uur
Universiteit Maastricht, Auditorium, Oxfordlaan 55, 6229 EV Maastricht
Voeg toe:

Lezing door James McClelland, laureaat van de C.L. de Carvalho-Heinekenprijs voor de Cognitiewetenschap 2014, over interactieve processen in taal en perceptie.

De lezing is vrij toegankelijk, maar het aantal plaatsen is beperkt (maximaal 70). Aanmelden is niet nodig.

Abstract Lecture James McClelland

Since the 1950's, it has been known that the medial temporal lobes in the brain play a special role in learning an memory.  These findings have led, through the work and thinking of David Marr and many others, to a theory of the roles of hippocampus and neocortex in memory called the complementary learning systems theory (McClelland, McNaughton, & O’Reilly, 1995). Our theory postulated two distinct learning systems, one in the medial temporal lobes that supports the rapid learning of arbitrary new information, and one in the neocortex and other structures that supports the gradual discovery of structured representations that encode knowledge of the natural and man-made world, as well as the knowledge underlying cognitive skills and the knowledge underlying language and communication.  In this talk, I consider how these two distinct learning systems function separately and together to allow us to learn and to remember.  First, I discuss their complementary roles in memory formation, describing how the MTL system provides a way of allowing the rapid acquisition of arbitrary new information while avoiding interference of such learning with the structured knowledge systems in the neocortex.  After briefly considering how these systems can work together in memory recall, I then examine evidence from recent studies showing that new information can sometimes be integrated rapidly into the neocortex, challenging our theory as previously presented. I present new simulations based on our theory, showing that new information that is consistent with knowledge previously acquired by a cortex-like artificial neural network can be learned rapidly without interfering with existing knowledge. These results match the pattern observed in the recent studies, and provide a mechanism for understanding when and how rapid integration of new information can occur.