Research and Education Centre in Adaptive Systems
by Miroslav Kárny
In the eighties, adaptive control and adaptive signal processing, and also briefly adaptive systems, made very fashionable research areas. Nowadays, any further development of particular techniques is performed mostly under other labels, like intelligent or predictive control (due to the fashion and policy of grant agencies). At the same time, the close connection to the classical statistical decision making, coined by A.A. Feldbaum in the sixties, has been revived. The available computer power, progress in solution of various particular tasks as well as better understanding of the addressed problems have opened new perspectives in adaptive systems understanding.
Decision making under uncertainty is quite a multi-disciplinary research branch both with respect to theory and application. As stated above, adaptive systems can be viewed as its engineering version with a stress on sequential problems. Their efficient research and teaching have to cross boundaries of traditional fields as well as artificial institutional boundaries. Various research institutes and universities dealt with then and they naturally focuse on the problems related to that particular institution (technology, economy, medicine,...) and as such are more applied in their perspective. This observation has led to the establishment of the Research and Education Centre in Adaptive Systems (RECiAS). It is supported by the Grant Agency of the Czech Republic. It intends to serve as an informal mediator in tailoring the research and education capacity to the problem at hand and to the student to be educated.
RECiAS, is a joint initiative of the Department of Adaptive Systems at UTIA and co-operating universities (Czech Technical University, Prague, Charles University, University of West Bohemia, PlzeÀ, Reading University), research institutes (Computer and Automation Institute, Budapest) and application oriented partners (Compureg, PlzeÀ, Glass Service, Vsetín). The principal aim of this Centre is to support education of top experts in the field of Adaptive Systems.
RECiAS offers a new teaching framework with the following main priorities
- learning by doing research and development is the dominant approach adopted
- crossing boundaries between departments, universities and research institutes are intentionally crossed both with respect to student acquisition; their specialised lectures as well as their supervisor.
The centre serves for (pre-dominantly postgraduate) education of theoreticians and practitioners and will be counted successful only if it produces top experts.
This aim manifests itself in:
- completion of the existing teaching means by lecture notes unifying the underlying theory, experimental laboratory, demo examples and sets of typical real data
- incorporation of students into running newly proposed research projects so that reviews, research reports, diploma and PhD theses will be created (areas: fault detection, decision support in the case of nuclear accident, biomedicine oriented modelling, estimation of relationships in a complex transportation systems, etc.)
- incorporation of students into running and newly proposed applied projects (early warning system RODOS, data processing in nuclear medicine, radon dynamics, fast prototyping of parallel systems, control design of rolling systems, operator support, traffic control, etc).
An Example of a Problem
Accumulation kinetics of 131I used for thyroid cancer treatment exhibited unacceptable discrepancies between the estimation (diagnostic) and application (therapeutic) phase. A careful Bayesian analysis showed that under-modelling is responsible for this. The decision making of medical doctors on applied therapeutic activity will be substantially simplified by this result.The figure shows data (*) measured since the application of 131I, their fit (o) when the classical mono-exponential model, characterised by the time constant (tTef), is used and fit (x) by an alternative model (tAlt). The line ñ shows predictive ability of the former model (dTef) when data from the accumulation beginning are available (the period coincides with the length of diagnostics). The improvements of predictions (-.) for the alternative (dAlt) is obvious.
Links:
RECiAS, partners and specifics projects: http://www.utia.cas.cz/user_data/scientific/AS_dept/RECIAS/
Please contact:
Miroslav Kárny - CRCIM
Tel: +420 2 688 3421
E-mail: school@utia.cas.cz
