Abstract:
Reinforcement learning (RL) has demonstrated an ability to succeed in a few arduous tasks, emerging as a general framework for decision making in neuroscience and robotics. However, engineering approaches to solve this problem differ in many respects with those in which the brain implements. This brings up the question as to how the human brain develops an ability to handle a wide variety of tasks and to learn from only few observations. This talk introduces our research team’s twofold approach to better understanding the nature of human RL.
The first part of the talk focuses on understanding the prefrontal-striatal circuitry for meta-RL. The theoretical idea explains how the human brain resolves major tradeoff issues: performance-efficiency, speed-accuracy, and explorationexploitation. The second part of the talk outlines a more pragmatic approach to improving optimality of human RL. A detailed insight into these issues not only permits advances in a reinforcement learning theory, but also helps us understand the nature of human intelligence on a deeper level.
Sponsored by the NYU-ECNU Institute of Brain and Cognitive Science at NYU Shanghai