A Simple-Units Complex-Structure Neural Network model of the Basal Ganglia to simulate reinforcement learning tasks

• Published in: Procedia computer science Vol. 192; pp. 281 - 290
• Main Authors: Drapała, Jarosław, Frydecka, Dorota, Świątek, Jerzy
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 2021
• Subjects: Basal Ganglia model; Go an NoGo learning; instructed probabilistic selection task; learning form rewards; punishments; Go an NoGo learning; learning form rewards; instructed probabilistic selection task; Basal Ganglia model; punishments
• DOI: 10.1016/j.procs.2021.08.029
• ISSN: 1877-0509

In the computation psychiatry field, the reinforcement learning tasks aim at measuring a subject’s sensitivity to rewards and punishments. We aim at providing a mechanistic account of behavioral data of participants undergoing reinforcement learning tasks in Wroclaw Medical University by quantitatively reproduce the observed tendencies through computer simulations of the developed Simple-Units Complex-Structure Neural Network. The network mimics the core properties of the Basal Ganglia which is a group of subcortical nuclei present in the brain responsible for motor control and learning from rewards and punishments. We demonstrate the performance of the proposed network on three reinforcement learning tasks: probabilistic selection task, probabilistic reversal task, and instructed version of the probabilistic learning task. Our simulations show that the network can express the behavior observed in studies of human subjects performing reinforcement learning tasks.