Dopamine: a moment-by-moment teaching signal connecting AI and brain
Prof. HyungGoo Kim
March 11(Thu) - March 11(Thu), 2021
12PM
Room 86314 (Multi-purpose room), N center/ ZOOM (728-142-6028)
Neuro@noon Seminar
Date: 12:00 PM, Thursday, Mar 11th
Place: #86314 & ZOOM
이번 Neuro@Noon은 온라인과 오프라인 동시에 진행이 됩니다. 시간이 가능하신 분들은 오프라인으로 참석해주시면 감사하겠습니다.
1. Offline 참여 방법
N센터 3층-86314호 (Multi-Purpose room)
2. Online ZOOM 참여 방법
아래의 링크를 클릭 하거나,
ZOOM 홈페이지 혹은 어플에서 아래의 회의 ID를 입력하면 세미나에 참여가 가능하게 됩니다.
과학 뇌이(가) 예약된 Zoom 회의에 귀하를 초대합니다.
https://us02web.zoom.us/j/7281426028
회의 ID: 728 142 6028
Speaker: Prof. HyungGoo Kim
Department of Global Biomedical Engineering, Sungkyunkwan University, Korea
Title: Dopamine: a moment-by-moment teaching signal connecting AI and brain
Abstract:
To survive and thrive in uncertain environments, humans and animals adjust their behaviors to maximize reward. Reinforcement learning (RL), originally a machine-learning algorithm implementing how an agent takes actions to maximize reward in a virtual world, provides a theoretic explanation to real-life
adaptive behaviors. However, the biological mechanisms by which the brain implements RL have been elusive. Transient dopamine activity has been thought as a teaching signal (reward prediction error) to update reward prediction. It was unclear, however, what the slowly fluctuating dopamine activity during spatial navigation represents. Directly testing fluctuating dopamine in RL theory would require instant repositioning of the animal to a new location. I took advantage of visual virtual reality and measured dopamine activity while mice were ‘teleported’ during navigation. Using this system, I demonstrated that dopamine neurons compute moment-by-moment reward prediction error, providing for the first time a biological basis to a key theoretic component of RL implementation in the brain.