Search Papers | Poster Sessions | All Posters
Poster B15 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
Neural dynamics of reversal learning in the prefrontal cortex
Christopher Kim1 (), Carson Chow1, Bruno Averbeck2; 1Laboratory of Biological Modeling, NIDDK/NIH, 2Laboratory of Neuropsychology, NIMH/NIH
We used a reversal learning task with probabilistic reward to investigate how a neural network accumulates evidence across multiple trials to reverse its decision. We analyzed prefrontal cortex activity in monkeys performing the task and recurrent neural networks trained to learn the behavioral strategies of monkeys. We found substantial neural dynamics across the time span of a trial in the subspace that encodes reversal probability. This suggested that the standard attractor model for evidence accumulation, in which network states do not deviate strongly from attractor states, does not explain the observed neural activity. We found that reward outcomes affected the entire reversal probability trajectories systematically. The reversal probability trajectories across trials had temporally stable ordering, and the reversal trial was decodable over a wide time span. These findings show that, when performing a task that requires intervening behavior, reversal probability activity across trials is encoded in dynamic neural trajectories, allowing temporally flexible representation of decision-related evidence.
Keywords: Reversal learning Recurrent neural network