![[picture]](figs/john-small.gif)
John van den Elst
John van den Elst
ABB Corporate Research John van den Elst Gideonsbergsgatan 2 S-72178 Västerås Sweden Tel : +46 21 32 32 48 Fax : +46 21 14 21 90 Email : john.van-den-elst@secrc.abb.se
Several scientific disciplines have evolved in such a way that they depend heavily on predefined complex methods and techniques. These methods and techniques are developed by specialists, and solve a subpart of the overall problem. In the domains of image processing, scientific computing, signal processing and automatic control, they are often implemented in computer programs and stocked in a library of programs. Solving a new problem in such a domain requires both extensive experience in the programs contained in the library, and detailed knowledge of the overall task for which they are applied. Frequently those who need such programs have knowledge of the overall task, but not the experience with the programs themselves, and thus need assistance in the choice, practical use and result interpretation of the programs. A powerful solution to this problem relies on a tight integration, within the application domain, of the knowledge associated to programs and to entities handled by the programs, as well as knowledge about how to select the adequate programs, how to schedule them, how to run them and eventually how to control their execution and optimise the results for a particular context. We call this the problem of program supervision.
The PhD thesis work concerns research to how AI techniques can be used in this domain, resulting in a knowledge model for program supervision. Theoretical issues concerning knowledge modelling and acquisition are based on the work done at the University of Amsterdam. They involve the definition of the knowledge representation formalisms, together with the way these should be used. The necessary formalisms can be found in the fields of reuse in software engineering, problem solving, expert systems, meta-architectures, planning, and knowledge acquisition. These formalisms are the basis for the planning of programs to reach a certain predefined goal, and the dynamic execution of different series of programs (initialisation, evaluation, adaption). Research in the field of program supervision is one of the main axes of the work performed in the ORION project at INRIA. There we already have obtained first experiences with program supervision, especially in the field of image processing (system OCAPI). For the PhD thesis work, the domain of autonomous vision systems is used, applied to medical image processing, and object detection in road scenes. The proposed solutions are primarily developed for this application domain, but are still kept as general as possible.