The information geometry of flexible decision making

Poster Session B - Sunday, March 30, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Anne Marazita¹ (anne.marazita@pennmedicine.upenn.edu), Lowell W. Thompson¹, Juvenal Bosulu¹, Jean Zweigle¹, Joshua I. Gold¹; ¹University of Pennsylvania

Adaptive behavior depends on the flexible use of new information. This flexibility involves balancing two seemingly contradictory goals: 1) increasing the certainty of our beliefs by prioritizing information that reinforces those beliefs, and 2) increasing the accuracy of our beliefs by prioritizing novel and potentially contradictory information. Here, we present a new theoretical framework showing that this balance arises naturally from information geometry and apply the framework to understand how the brain uses new information to guide behavior on a flexible decision-making task. We trained two monkeys to decide which of two visual targets will be rewarded, based on: (1) a partially predictive visual cue; and (2) a learned expectation about the rate at which the rewarded target switched from trial to trial. In low switch-rate blocks (P(switch)=0.05), visual cues were most useful when they indicated a switch. In high switch-rate blocks (P(switch)=0.5), visual cues were most useful when they established new, strong beliefs. We recorded from individual neurons in the dorsolateral prefrontal cortex (dlPFC) of one monkey (N = 95) performing the task. The monkeys flexibly balanced observations and expectations depending on the switch rate. This flexibility was reflected in the dlPFC at the single-neuron and population level. Consistent with theoretical predictions, visual-evoked responses were stronger in the high-switch condition. Conversely, visual-evoked responses were lower, and delay-period activity was higher, in the low-switch condition. These findings provide new insights into how the brain makes flexible use of information to update beliefs that guide behavior.

Topic Area: THINKING: Decision making