Theory and Implementation of Multi-Context Systems Containing Logical and Sub-Symbolic Contexts of Reasoning.
Tarek R. Besold · OPUS FAU (Kooperativer Bibliotheksverbund Berlin-Brandenburg (KOBV), on behalf of the Universitätsbibliothek Erlangen-Nürnberg) · 2010
In the introductory part, we give a brief overview of the state of the art concerning multi-context systems (MCS), giving some recent examples from the literature, as well as lining out advantages and disadvantages of the different approaches. Then we propose an extension of the heterogeneous multi-context reasoning framework by G. Brewka and T. Eiter, which, in addition to logical contexts of reasoning, also incorporates sub-symbolic contexts of reasoning. The main findings concerning this topic are a simple extension of the concept of bridge rules to the sub-symbolic case and the concept of bridge rule models that allows for a straightforward enumeration of all equilibria of multi-context systems. Also a very basic, yet applicable algorithm for solving this task is presented, and our approach is illustrated with two examples from the fields of text and image classification. Moreover, after some theoretical considerations containing refinements and an expansion of the basic algorithm, we present a proof of concept implementation of an MCS, already integrating different techniques for reducing computational complexity. These techniques have been developed for this very purpose and are described and analyzed as well. The main ideas are a formalism to impose constraints on bridge rules, allowing to state dependencies between different bridge rules or sets of bridge rules, and the concept of conflicting bridge rules, which allows for the application of pruning techniques within the possible set of equilibria of the MCS. Again we illustrate our approach with three examples taken from different domains of application, having a closer look at a special purpose application of multi-context systems made for museum data completion and consistency checking. Finally possible future prospects and extensions of MCS are sketched, presenting inter alia the notion of generalized bridge rules and bridge rule inference. To conclude the thesis a comparison of our work with similar or related approaches is given.