S. Bairakdar:
"Local Optimization for Multi-Context Systems with Constraint Pushing";
Supervisor: T. Eiter, M. Fink, T. Krennwallner; Institut für Informationssysteme, 2011; final examination: 2011-04-13.

Multi-Context Systems (MCS) are formalisms that enable the inter-linkage of single knowledge bases called contexts, via bridge rules. In this thesis, we focus on Brewka and Eiter-style heterogeneous nonmonotonic MCS and develop a novel distributed algorithm for computing the equilibria of such MCS. We examine previous approaches for distributed MCS evaluation that have been implemented as part of the DMCS system. Moreover, the notion of association rules and the process of association rule extraction from the data mining field are recalled. None of the available techniques addressed the issue of local optimization within a distributed evaluation, which is the motivation behind our work. Our approach for local constraint pushing (DMCS-SLIM), relies on the coupling of some optimization techniques for distributed MCS evaluation with local association rules extraction. We prove DMCS-SLIM to be sound and complete. Furthermore, we present a prototypical implementation, which is used for empirical evaluation. We performed exhaustive set of experiments against several runtime parameters of our system as well as comparisons with existing approaches. We observed that our approach has potential to surpass the current approaches.

Project Head Thomas Eiter:
Modulare HEX-Programme

Project Head Michael Fink:
Inconsistency Management for Knowledge-Integration Systems