Doctoral seminars

< back

Tuesday November 13 at 14h00 in Celestijnenlaan 200A, room 03.14 (Cafetaria) Coherent Composition of Distributed Knowledge-Bases through Abduction: an experiment with the A-system by Bert Van Nuffelen

Coherent Composition of Distributed Knowledge-Bases through Abduction: an experiment with the A-system

In the past 50 years, there has been a boost in acquired knowledge by the human kind.
During the same period, the possibilties to store this knowlegde have been tremendeously growing. Many researchers and companies have put, and do this still today, a lot of effort in finding good techniques to manage information. Currently, one of the typical approaches is to distribute the knowledge over several fysical independent databanks which share the same knowledge skeleton. Each of them contains different pieces of the stored information. This approach creates one big issue: how to keep the information consistent when it is scattered around? Or another situation: What to do when inconsistencies arise?

In this presentation we focus on the last question. To decribe this we use a subset of the expressive language ID-logic, namely the part which is mappable into the Abductive Logic Programming framework. This enables us to get a high level meta theory of the problem which can be experimentallly verified. At the same time this approach puts very few restrictions on the skeleton description language. We start from a simple base case and explore preliminary some extensions, which capture more complex skeletons.

This obtained descriptions has been used in some small experiments with a new abductive system, the A-system. The A-system has been recently developed and can be used to solve highly complex tasks like planning and scheduling. Based on the insights and features of two earlier systems: SLDNFA and ACLP, the A-system is build with a more advanced search control. Instead of sticking to a depth first strategy, the A-system performs a propagation phase after each choice and tries to select the best possible next choice. As consequence the system becomes more robust and obtains a good overall performance.

This presentation combines two papers which are (or will be) presented at IJCAI'01 and LPAR'01.