McGovern Institute Special Seminar: Rajesh Rao
Wednesday, February 14, 2024 at 12:30pm to 2:00pm
Building 46, 3189
43 VASSAR ST, Cambridge, MA 02139
Rajesh P. N. Rao is the CJ and Elizabeth Hwang Professor in the Paul G. Allen School of Computer Science and Engineering and Department of Electrical and Computer Engineering at the University of Washington (UW), Seattle. He is also the co-Director of the Center for Neurotechnology (CNT), Adjunct Professor in the Bioengineering department, and faculty member in the Neuroscience Graduate Program at UW. He directs the Neural Systems Laboratory located in the Paul G. Allen Center for Computer Science and Engineering. He is an IEEE Fellow and the recipient of a Guggenheim Fellowship, a Fulbright Scholar award, an NSF CAREER award, an ONR Young Investigator Award, a Sloan Faculty Fellowship, and a David and Lucile Packard Fellowship. His research interests span computational neuroscience, brain-computer interfaces, and artificial intelligence. His other interests include the ancient Indus script and classical Indian paintings.
A sensory-motor theory of the neocortex based on active predictive coding
Recent neurophysiological experiments indicate that almost all cortical areas, even those traditionally labelled as primary sensory cortices, are modulated by upcoming actions. Parallel evidence from neuroanatomical studies points to major outputs from layer 5 neurons across cortical areas to subcortical motor centers. To account for these findings, we propose that the neocortex implements active predictive coding (APC), a form of predictive coding that combines actions and hierarchical sensory-motor dynamics. We provide examples from simulations illustrating how the same APC architecture can solve problems that seem very different from each other: (1) how do we recognize an object and its parts using eye movements? (2) why does perception seem stable despite eye movements? (3) how do we learn compositional representations, e.g., part-whole hierarchies, and nested reference frames? (4) how do we plan actions in a complex domain by composing sequences of sub-goals and simpler actions, and (5) how do we form episodic memories of our sensory-motor experiences and learn abstract concepts such as a family tree? The APC model suggests computational roles for interlaminar connections within a cortical area as well as possible roles for cortico-cortical and cortico-subcortical connections.