FriSem

Robert Guangyu Yang,  Assistant Professor in the Department of Brain and Cognitive Sciences and EECS at MIT

Title: Towards integrative neural network models of working memory

Abstract: Working memory is one of the oldest problems in cognitive psychology. Some of its key features have been known for decades: remarkable flexibility and limited capacity. Yet we still don't know how working memory can be so flexible and why there exists a strong capacity limit. In this talk, I'd like to discuss some ongoing work in my group building integrative neural network models of working memory, and our attempts at understanding their flexibility and capacity. Here by "integrative" I mean the model consists of at least two systems: a visual system that processes pixel inputs and a cognitive system that supports working memory. By having a working memory model that deals with complex, high-dimensional inputs, we can assess their flexibility and capacity in new, exciting ways. We show the conditions under which these network models have flexibility and even capacity limits similar to humans and many other animals.