Ioan Doré LANDAU
Biography
Biography
Ioan Doré LANDAU is Emeritus Research Director at C.N.R.S.(National Centre for Scientific Research) since September 2003 and continues to collaborate with the GIPSA-LAB (CNRS/UGA),Grenoble.
His research interests encompass theory and applications in adaptive control, system identification, robust digital control and nonlinear systems. He has authored and co-authored over 400 papers on these subjects. He is the author of the books : Adaptive Control – The Model Reference Appoach ( Dekker 1979 translated in Chinese), System Identification and Control Design (Hermès 1993 ,Prentice Hall 1990). Commande de systèmes : conception, identification, mise en oeuvre (Hermes-Lavoisier, 2002, Translated in English, Romanian and Chinese) and co-author (with M. Tomizuka) of the book Adaptive Control – Theory and Practice (in Japanese -Ohm 1981) as well as co-author (with R. Lozano and M. M’Saad and A. Karimi) of the book Adaptive Control (Springer Verlag 1997, 2011). He edited and co-edited several books in french on the above topics including a series on «Mathematical Tools for Control, System Analysis and Signal Processing and Models » and one on « Adaptive Control ». He delivered a number of Plenary Talks at International Conferences including American Control Conference in Seattle in 95. He was the key note speaker at the European Control Conference in Bruxelles in 1997.
Dr. Landau received the Rufus Oldenburger Medal 2000 from the American Society of Mechanical Engineering which recognizes significant contributions to the field of automatic control for his pioneering contributions in adaptive control and system identification. He is “Doctor Honoris Causa” of the Faculté des Sciences, Université Catholique de Louvain-la- Neuve (2003) and of the University « Politechnica », Bucarest (2017). He was a R.Springer Professor at University of .California. Berkeley, Dept.of Mechanical Engineering in 1992. He received the price Monpetit from the French Academy of Science in 1991, the “ASME Honor Award (1981-84)” for his paper on adaptive control published in ASME Journal of Dynamical Systems Measurement and Control, the C.N.R.S. Silver Medal in 1982 and the Great Gold Medal at the Invention Exibition Vienna in 1968 for his patent on the variable frequency control of asynchronous motors. He was an IEEE-CSS “Distinguished Lecturer” for 2001-2003. At C.N.R.S. he was the Director of coordinated research programs: « Mathematical Tools and Models for Control, System Analysis and Signal Processing » from 1979 to 1982, « Adaptive Systems in Control and Signal Processing » from 1984 to 1988 and « Automatique » from 1988 to 1996. He was also Director of the Laboratoire d’Automatique de Grenoble from 1987 to 1990. Dr. Landau was the General Chairman of the first European Control Conference organized in Grenoble in 1991. He was one of the founders and the first President of the European Community Control Association (ECCA) from 1991 to 1993 (now EUCA) and he was Editor in Chief of the European Journal of Control (a publication of the European Union Control Association) from 1994 to end 2002.
Accelerating adaptation/learning/optimization algorithms
A today challenge
I.D. Landau
Emeritus Research Director at CNRS, GIPSA-LAB, Grenoble
With the becoming of neural networks, real-time identification, adaptive control and optimization techniques, there was in the last fifteen years an explosion of the number of parameter adaptation/learning/optimization algorithms (PALOA) proposed from different points of view. The underlying objective being the acceleration of the adaptation /learning/optimization process. The field became a kind of “fiddler’s paradise”. In most of the cases, only a qualitative analysis of their behavior has been provided. Parameter adaptation/ learning/optimization algorithms are nonlinear time varying dynamical systems with an inherent feedback structure for which stability is a key issue. Passivity approach appears to be a basic and efficient tool for understanding, analyzing and synthesizing the algorithms. All the algorithms proposed can be viewed as extensions of the basic gradient algorithm. The key point is the introduction of the concept of “dynamic adaptation gain” (DAG). The basic adaptation gain/learning rate is replaced by an embedded poles/zeros (ARMA) filter of any order, which should be characterized by a positive real transfer function. The key question is how to select the coefficients of this filter, on one hand, to guarantee the stability of the adaptation /learning/optimization process and on the other hand to accelerate the adaptation /learning/optimization process. Solutions to these problems will be presented. It will be shown that many popular algorithms (Conjugate gradient, Nesterov, Averaged gradient, Momentum back propagation, Integral + proportional + derivative) are particular cases of the structure of PALOA introduced above. A particular form of the DAG called ARIMA2 will be presented. The impact of the dynamic adaptation gain upon the performance of the PALA algorithms will be illustrated by comparative performance evaluation of various algorithms on relevant simulation examples and real time experiments on an adaptive active noise attenuation system.