laasri@crin.crin.fr (Hassan LAASRI) (07/07/88)
Originally, a blackboard-based system designed an AI system which is composed of a set of independent computational agents known as knowledge sources (KSs) interacting and communicating via a global database called the blackboard, under the management of a controller that can be also an AI system. Recently, a proliferation of "distributed blackboard systems" (or distributed control in blackboard systems) definitions were employed in knowledge-based systems such as the UMass SCHEMA system or the CMU Navlab system...In these systems, one read that, because the controller component can be a system bottleneck, the systems designers prefer to distribute the control between the KSs, so each KS is able to decide when and how it can contribute to the problem solving process. Unfortunately there is not enough documentations about their controllers. We would like to have some highlights on these questions : 1. In a distributed control model, once two or more KSs are triggered to activate, how do they negotiate between them to solve the conflict problem of the blackboard accesses, i.e. how they find the best order in which they will be executed ? do they use the contract net model or some thing like that ? 2. Does each KS evaluate its contributions to the problem solving evolvement and return a priority value, then the system executes the KS with the highest priority ? in this case, this control type will be only a subset of the original "centralised" control model used in systems such HEARSAY-II, BB-1, HASP/SIAP, ATOME, GBB,... P.S. We have decided to put this demand in this News group, after trying to contact the concerned researchers. Unfortunately, we didn't have any response to our requests. ----------------------------------------------------- Mr. Hassan LAASRI and Miss. Brigitte MAITRE CRIN/INRIA-Lorraine Pattern Recognition and Artificial Intelligence Lab. Blackboard Group Campus Scientifique - B.P. 239 54506 VANDOEUVRE-LES-NANCY CEDEX FRANCE E-mail : laasri@crin.crin.fr or maitre@crin.crin.fr Phone : (+33) 83-91-20-00 Post 30-05 -----------------------------------------------------