Tachmazidis, Ilias (2015) Large-scale Reasoning with Nonmonotonic and Imperfect Knowledge Through Mass Parallelization. Doctoral thesis, University of Huddersfield.
- Accepted Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.
Download (828kB) | Preview
Due to the recent explosion of available data coming from the Web, sensor readings, social media, government authorities and scientific databases, both academia and industry have increased their interest in utilizing this knowledge. Processing huge amounts of data introduces several scientific and technological challenges, and creates new opportunities. Existing works on large-scale reasoning through mass parallelization (namely parallelization based on utilizing a large number of processing units) concentrated on monotonic reasoning, which can process only consistent datasets. The question arises whether and how mass parallelization can be applied to reasoning with huge amounts of imperfect (e.g. inconsistent, incomplete) information. Potential scenarios involving such imperfect data and knowledge include ontology evolution, ontology repair and smart city applications combining a variety of heterogeneous data sources. In this thesis, we overcome the limitations of monotonic reasoning, by studying several nonmonotonic logics that have the ability to handle imperfect knowledge, and it is shown that large-scale reasoning is indeed achievable for such complex knowledge structures. This work is mainly focused on adapting existing methods, thus ensuring that the proposed solutions are parallel and scalable. Initially, preliminaries and literature review are presented in order to introduce the reader to basic background and the state-of-the-art considering large-scale reasoning. Subsequently, each chapter presents an approach for large-scale reasoning over a given logic. Large-scale reasoning over defeasible logic is supported allowing conflict resolution by prioritizing the superiority among rules in the rule set. A solution for stratified semantics is presented where rules may contain both positive and negative subgoals, thus allowing reasoning over missing information in a given dataset. The approach for stratified semantics is generalized in order to fully support the well-founded semantics, where recursion through negation is allowed. Finally, conclusion includes observations from a preliminary investigation on a restricted form of answer set programming, a generic evaluation framework for large-scale reasoning, a discussion of the main findings of this work, and opportunities for future work.
|Item Type:||Thesis (Doctoral)|
|Subjects:||Q Science > Q Science (General)
T Technology > T Technology (General)
|Schools:||School of Computing and Engineering|
|Depositing User:||Elizabeth Boulton|
|Date Deposited:||26 Jan 2016 09:47|
|Last Modified:||02 Dec 2016 04:41|
Downloads per month over past year
Repository Staff Only: item control page