Event Details:
Professor Surya Ganguli will begin the session with a brief update on the state of the Center and an invitation to newcomers to join in the Center's activities. Those wishing to join are invited to stay for dinner to meet Surya and learn more about MBC. Newcomers are also invited to become affiliates by requesting the affiliate signup form from lehope@stanford.edu
Presentation by Michael Waskom: "Adaptive engagement of cognitive control reflects a predictive model of task context"Abstract: Many decisions require a flexible, context-dependent mapping from sensory evidence to motor response. Although cognitive control can be used to enable this flexibility, control processes are capacity limited and their engagement is costly. It remains unknown how the brain balances these tradeoffs to optimize behavior in dynamic environments. In this talk I will introduce a novel perspective that extends theories about the importance of predictive models in neural computation. To explore this idea, we scanned human participants using fMRI while they performed a context-dependent perceptual decision task. On each trial, participants were cued to judge either the direction of motion or the dominant color of a bivalent random dot stimulus. Over the course of the experiment, we parametrically manipulated the relative frequency of motion and color trials without indicating to the participants which was more likely. We then developed a Bayesian model that learned these probabilities and allowed for predictions about the likely task context. Our results show that the engagement of cognitive control can be explained by a measure of context prediction error derived from this model. Behaviorally, decisions required longer deliberation when the context was unexpected. Frontoparietal regions also exhibited a larger evoked response on these trials, which increased the discriminability of distributed representations of task context. These results emphasize the importance of adaptive learning mechanisms in managing tradeoffs inherent to high-level cognitive processes.
RSVP REQUIRED: Email lehope@stanford.edu SEMINAR: Sloan Hall, Math Bldg 380, Room 380-C, lower level courtyard side, at 5:15pm DINNER FOLLOWING SEMINAR: Courtyard outside of Jordan Hall, Bldg 420, Room 050