WHO:	BCS Lunch Talk: Jeong-Woo Sohn , BCS Graduate Student
TOPIC:	Reinforcement Learning and the prefrontal cortex
ABSTRACT:	Animals interact with their environment and try to discover the optimal sequence of actions for their survival. According to reinforcement learning theory, optimal sequences of actions can be determined by maximizing the estimate of temporally discounted sum of future rewards, a quantity referred to as a value function. Typically, the value function is evaluated relative to various combinations of environmental states and actions of the animal. Therefore, an optimal sequence of actions can be generated by simply choosing at each time step an action that maximizes the value function. To understand whether and how reinforcement learning algorithms are implemented in the human brain, we developed a non-human primate model of reinforcement learning based on a serial reaction time (SRT) task. The results show that the behavioral performance is systematically influenced by the value function. Additional factors such as distance to the rewarded location also influenced the performance, suggesting that they might be used as heuristic measures of value functions. Preliminary neurophysiological recordings show that value functions might be represented in the prefrontal cortex.
WHEN:	10/29/2003 12:00:00 PM
WHERE:	Meliora 269

Events Homepage