Department News
A Unified Approach of Adaptive Nonlinear Systems in Complex Domain
Seminar Date
2006-02-02
Author
빈종훈
Date
2006-01-20
Views
1878
1. 제 목 : A Unified Approach of Adaptive Nonlinear Systems in Complex Domain
2. 연 사 : 김 태 환 , Lead Artificial Intelligence Scientist, The MITRE Corporation
3. 일 시 : 2006년 2월 2일(목요일) 오후 3:00-5:00
4. 장 소 : 서울대학교 301동 301호
5. 내 용 :
This seminar discusses the long-standing problem of how to build a robust on-line adaptive system. This important subject is rooted in adaptive signal processing (i.e., radar, sonar, communications, and machine learning) where Least Mean Square (LMS), Kalman filter, and neural networks are the essential tools These methods all suffer from modeling and/or learning error while the desire for faster on-line convergence is ever-increasing. Even more challenging, especially for neural networks, is the need to use complex-valued data as calculus in complex domain becomes more intricate and all analytic (differentiable) functions possess singularity (unbounded discontinuity).
To provide the cure for robustness and speedy on-line performance, a unified view of the adaptive system is introduced first. Then, it is shown that the complex domain approach is a blessing, not a curse. Application examples in system identification, communication, and navigation are provided.
6. 약 력 :
1997-Present The MITRE Corporation, McLean, VA
2002 Ph.D., Electrical Engineering, University of Maryland
1990-1997 NASA Goddard Space Flight Center, MD
(Stanford Telecommunication Inc.)
1987-1990 GSI, Annapolis, MD
1984-1986 M.S., Computer Science, University of S. Carolina
1978-1984 B.S., Mathematics, Seoul National University
7. 문 의 : 기계항공공학부 기 창 돈 교수 (☏ 880-1912)
2. 연 사 : 김 태 환 , Lead Artificial Intelligence Scientist, The MITRE Corporation
3. 일 시 : 2006년 2월 2일(목요일) 오후 3:00-5:00
4. 장 소 : 서울대학교 301동 301호
5. 내 용 :
This seminar discusses the long-standing problem of how to build a robust on-line adaptive system. This important subject is rooted in adaptive signal processing (i.e., radar, sonar, communications, and machine learning) where Least Mean Square (LMS), Kalman filter, and neural networks are the essential tools These methods all suffer from modeling and/or learning error while the desire for faster on-line convergence is ever-increasing. Even more challenging, especially for neural networks, is the need to use complex-valued data as calculus in complex domain becomes more intricate and all analytic (differentiable) functions possess singularity (unbounded discontinuity).
To provide the cure for robustness and speedy on-line performance, a unified view of the adaptive system is introduced first. Then, it is shown that the complex domain approach is a blessing, not a curse. Application examples in system identification, communication, and navigation are provided.
6. 약 력 :
1997-Present The MITRE Corporation, McLean, VA
2002 Ph.D., Electrical Engineering, University of Maryland
1990-1997 NASA Goddard Space Flight Center, MD
(Stanford Telecommunication Inc.)
1987-1990 GSI, Annapolis, MD
1984-1986 M.S., Computer Science, University of S. Carolina
1978-1984 B.S., Mathematics, Seoul National University
7. 문 의 : 기계항공공학부 기 창 돈 교수 (☏ 880-1912)
