We cordially invite you to a seminar on reasoning over ontologies and data, organized at the VU University Amsterdam. The seminar is held in the context of the PhD defence of Szymon Klarman from the Knowledge Representation and Reasoning group.
Time: Wednesday, February 27, 10:15 – 13:00
Place: VU University Amsterdam, W&N building, room WN-S631 (De Boelelaan 1081a).
10:15 Welcome & Coffee
10:30 – 11:15 Shape and Evolve Living knowledge – a case on procedural and ontological knowledge.
Chiara Ghidini, Data and Knowledge Management group, FBK Trento.
11:20 – 12:05 Non-Uniform Data Complexity in Ontology-Based Data Access with Description Logics.
Carsten Lutz, Theory of Artificial Intelligence group, University of Bremen.
12:10 – 12:55 Probabilistic Reasoning for Web-Scale Information Extraction.
Heiner Stuckenschmidt, Data and Web Science group, University of Mannheim.
Chiara Ghidini: Shape and Evolve Living knowledge – a case on procedural and ontological knowledge
The ability to effectively manage business processes is fundamental to ensure the efficiency of complex organizations, and a key step towards the achievement of this ability is the explicit representation of static and dynamic aspects of the organization in the form of conceptual models.
Shaping and maintaining these conceptual representations, and represent them in appropriate logical formalisms still presents many open challenges. The aim of the newly launched SHELL (Shape and Evolve Living knowledge) project is to tackle key interdisciplinary challeges in the fields of (i) the shaping of conceptual models of an organisation, (ii) their representation in appropriate formalisms, and (iii) their co-evolution and adaptation w.r.t. data.
In this talk I will provide: (i) an overview of the SHELL project, (ii) an illustration of our approach for the representation and verification of structural properties of integrated BPMN business processes and OWL ontologies, and (iii) hints on some on-going work towards the connection of this representation with data coming from real process executions.
Carsten Lutz: Non-Uniform Data Complexity in Ontology-Based Data Access with Description Logics
The idea of ontology-based data access (OBDA) is that an ontology (a logical theory) gives a semantics to the predicates used in a database, thus allowing more complete answers to queries, enriching the vocabulary available for querying, and mediating between data sources with different vocabularies. In this presentation, I will discuss OBDA with ontologies formulated in description logics (DLs) and advocate a novel approach to studying the data complexity of query answering in this context. The approach is non-uniform in the sense that individual ontologies are considered instead of all ontologies that can be formulated in a given DL. It allows us to ask rather fine-grained questions about the data complexity of DLs, such as: given a DL L, how can one characterize the ontologies for which query answering is in PTime or FO-rewritable? Is there a dichotomy between being in PTime and being coNP-hard? We provide several answers to such questions, some of which are based on a new connection between query answering w.r.t. DL ontologies and constraint satisfaction problems (CSPs) that allows us to transfer results from CSPs to DLs. We also identify a class of ontologies within the expressive DL ALCFI that enjoy PTime data complexity; the new class strictly extends the Horn fragment of ALCFI, which was was the largest known tractable fragment of ALCFI so far.
Heiner Stuckenschmidt: Probabilistic Reasoning for Web-Scale Information Extraction
Most of the current information extraction systems, therefore, provide a degree of confidence associated with each extracted statement. The higher this numerical value the more likely is it a-priori that the statement is indeed correct. While probabilistic knowledge bases provide a natural representational framework for this type of problem, probabilistic inference poses a computationally challenging problem. In our work, we want to distribute a sampling-based inference algorithm whose input is (a) a large set of statements with confidence values and (b) existing background knowledge, and whose output is a set of statements with a-posteriori probabilities. We propose the development and implementation of a distributed inference algorithm that has two separate processes running on the Hadoop platform. The first process constructs hypergraphs modeling the statements and their conflicts given known background knowledge. The second process runs Markov chains that sample consistent sets of statements from the conflict hypergraph ultimately computing the a-posteriori probabilities of all extracted statements.