반복영역 바로가기
주메뉴로 바로가기
좌측메뉴로 바로가기
본문으로 바로가기

Home News&ActivitySeminars

Seminars

프린트페이스북

(대학원생 마일리지 적용) 학과 세미나 안내 (7/12(금) 1시 30분, 프린스턴대 이동헌 교수 / Seminar Notice : July 12 at 13:30 pm, Dr. Donghun Lee, Princeton Univ.)

2019.07.10 15:47

교수님학생분들께,

  

안녕하세요.

산업시스템공학과 세미나가 다음과 같이 진행될 예정입니다.

 

 

1. 일시장소 : 7/12(금) 오후 1시 30분, 산업경영학동 2멀티미디어실(2501)

                      

2. 주제Max-Bias in Q-learning: Consequences and Corrections

           

3. 연사 : 이동헌 박사(Princeton University)

 

4. 언어: 한국어

          

           

많은관심과참석부탁드립니다.

감사합니다.

            

김 민 경 드림

      

 

 

Dear ISysE Professors and Students,

 

           

ISysE dept. office invites you to attend the following semiar.

                     

DATE & TIME: Friday, July 12 at 1:30 pm

VENUE: E2 Bld.,  #2501

TITLE: Max-Bias in Q-learning: Consequences and Corrections

SPEAKER : Donghun Lee, Princeton University

LANGUAGE: Korean

  

We look forward to your attendance and encourage you to forward this invitation to colleagues who may be interested in the topic.

 

With kind regards,

 

Min Kyugn Kim

 

 

 

## Abstract Q-learning is a classic reinforcement learning algorithm that influenced many modern algorithms found in successful AI applications.

Its asymptotic convergence properties are remarkable, but the practitioners should use caution as Q-learning contains systematic max-bias. This talk will illuminate the consequence of naive application of Q-learning and present bias-corrected Q-learning, using two different problems:

the game of Roulette, and the control problem of intelligent batteries in smart grids. 

## Bio - B.A. in Biochemistry from Columbia University - M.S. in Computational Biology from Carnegie Mellon University - Senior Software Engineer in Samsung Electronics - Ph.D. in Computer Science from Princeton University   
 

## Key Reference

D. Lee and W. B. Powell, "Bias-Corrected Q-Learning with Multistate Extension," in IEEE Transactions on Automatic Control. doi: 10.1109/TAC.2019.2912443 https://ieeexplore.ieee.org/document/8695133
 
## Contact: Prof. Woo Chang Kim (wkim@kaist.ac.kr)

List