Search Papers | Poster Sessions | All Posters
Poster A10 in Poster Session A - Tuesday, August 6, 2024, 4:15 – 6:15 pm, Johnson Ice Rink
Behavioral and neural evidence for dynamic model arbitration in dorsolateral prefrontal cortex
Jae Hyung Woo1 (), Michael Wang1, Ramon Bartolo2, Bruno Averbeck2, Alireza Soltani1; 1Dartmouth College, 2National Institute of Mental Health
One of the hallmarks of higher cognitive function is the ability to link outcomes to relevant features of the environment, while ignoring the irrelevant features. This is especially relevant for learning and decision making under uncertainty, where any features or attributes of a selected option can be predictive of rewards. It has been suggested that the brain tackles such uncertainty by running multiple internal models of the environment and arbitrating among them based on their reliability. To reveal the neural mechanisms underlying this dynamic arbitration process, we carried out high channel count recordings in dorsolateral prefrontal cortex (dlPFC) while monkeys performed a probabilistic reversal learning task with multiple layers of uncertainty. By fitting choice behavior with models based on reinforcement learning, we found evidence for dynamic, competitive interaction between stimulus-based and action-based learning strategies. dlPFC was involved in arbitration in two ways: (1) arbitration weight was represented in the activity of dlPFC neurons; (2) only the relevant information for the currently adopted strategy was encoded congruently as the monkey’s subsequent choice. These results suggest that dlPFC could be crucial for flexible arbitration between alternative models of the environment.
Keywords: decision-making reinforcement learning cognitive control dorsolateral prefrontal cortex