Kolloquium/Ringvorlesung des Graduiertenkolleg Wissensrepräsentation

UML class diagrams can be used as a semi-visual language for expressing a conceptual model of a domain. The UML suggests to use two metaconcepts, "aggregation" and "composition", for expressing part-whole relationships in a class diagram. We discuss their semantics from a foundational ontology point-of-view and make a number of proposals how to extend the UML in order to obtain a more satisfactory treatment of part-whole relationships.

UML is a standard semi-visual notation which has been used as a visual syntax for conceptual modeling and ontology representation. This presentation shows how Upper-level ontologies (such as GFO, and DOLCE) can be used to provide real-world semantics for UML conceptual modeling constructs (e.g. class, powertypes, attributes, associations, roles and states). Moreover, it addresses the use of OntoClean's ontology of universals (and its associated meta-properties) as concrete guidelines to constrain the structure of UML conceptual diagrams. Finally, it discusses the BWW approach (the most relevant related work in the literature) which is based on Mario Bunge's Formal ontology.

Logic programming has been introduced as programming in the Horn clause subset of first order logic. This view breaks down for the negation as failure inference rule. In this seminar, we investigate an alternative declarative reading of logic programming. It is well-known that Horn Logic Programming is suitable for representing positive inductive definitions. In fact, many prototypical Prolog programs (member, append, ..) are simple inductive definitions. The question is whether this view can be extended to the case of recursive logic programs with negation. A problem here is that currently, there is no standard universally accepted theory of non-monotone induction in mathematical logic. I will discuss two different forms of non-monotone induction and argue that the well-founded semantics of logic programming correspond to one of these, namely iterated induction. I will compare logic programming under the well-founded semantics with an existing logic of iterated induction and argue that the first provides a superior formalisation of iterated induction. So, this study provides new insight on the declarative reading of logic programming, and it provides also new insight in the nature of inductive definitions.

Those coming to nomonotonic reasoning for the first time can find the subject rather perplexing. The purpose of this talk is to show how some of the mystery may be taken out of the subject for the student and newcomer. We show how nonmonotonic reasoning arises naturally out of classical logic, provided certain misunderstandings are avoided and an old habit is put aside.

Indeed, there are logics that act as natural bridges between classical consequence and the main systems of nonmonotonic inference to be found in the literature. Like the nonmonotonic systems, these bridge logics are supraclassical, but like classical logic they are perfectly monotonic. As well as providing easy passage from one to the other, they have an interest of their own.

Metadata for the World Wide Web is important, but metadata for Peer-to-Peer (P2P) networks is absolutely crucial. In this talk I will give an overview of the open source project Edutella and related projects, which combine semantic web and peer-to-peer technologies in order to make distributed repositories possible and useful.

I will start by describing the main services of the Edutella network infrastructure and its architecture based on the exchange of RDF metadata, starting with the query service as one of the core services of Edutella. This service is based on the Edutella Common Data Model (ECDM) and the Edutella query exchange language (RDF-QEL-i), which allows us to implement distributed queries over the Edutella network.

I will then sketch a new P2P routing topology which minimizes broadcast traffic and distance between the peers in such a networks, which will be used as a super-peer backbone network for the Edutella system. I will further discuss possibilities for implementing personalization functionalities in the Edutella enviroment, which generalize adaptive hypermedia functionalities so far available only in closed environments.

What is input/output logic, and why do we need it? The motivation comes from the study of conditional directives, e.g. obligations, goals, ideals. We need an approach that does not presume that directives carry truth-values, and does not have the effect that conditions automatically emerge as goals. Input/output logic seeks to do this. It is not another non-classical or modal logic, based on a peculiar axiom system or possible-worlds construction. In a sense, it is not a new logic at all; but rather a new way of deploying good old classical logic in a manner that makes sense for directives.

Spaces with distance measures are useful in various fields of computer science: in geographic information systems they model geographic distances between points in space, on sets of proteins they model various kinds of similarities between proteins, etc. I will introduce a hierarchy of different languages for representation and reasoning about distance spaces, discuss their expressive power and algorithmic properties.

Planen hat in der Kuenstlichen Intelligenz seit jeher grosse Beachtung gefunden, und zu dieser Problemstellung sinde ueber die letzten Jahrzehnte kontinuierlich neue Beitraege geleistet worden. Logik-basierte Planungsansaetze haben dabei von den Anfaengen weg eine besondere Rolle gespielt. Das Interesse an logik-basierten Methoden zur Plangenerierung hat insbesonders durch gute Erfolge bei der Transformation in logische Erfuellbarkeitsprobleme zugenommen, die vor einigen Jahren erzielt worden sind. Dadurch ermutigt sind in der letzten Zeit wiederholt Vorschlaege praesentiert worden, wie Planungsprobleme durch die mir sogenannter Answer Set Programmierung, die ein Zweig der nichtmonotonen logischen Programmierung ist, deklarativ formuliert und geloest werden koennen. In Weiterverfolgung dieser Idee, nach der Plaene als Modelle eines logischen Programms generiert werden, ist das Planungssystem DLV^Kentwickelt worden. Dieses erlaubt in der genuinen Planungssprache K eine regel-basierte Beschreibung des Planungsproblems unter Zuhilfenenahme von Default-Negation und bietet Moeglichkeiten wie parallele Aktionsausfuehrung, nichtdeterministische Effekte der Aktionen und die Berechnung optimaler Plaene. Damit lassen sich auch teils vertrackte Plannungsprobleme verhaeltnismaessig einfach erfassen und loesen. Neben dem prinzipiellen Systemansatz, der auf Answer Set Programmierung aufbaut, werden Beispiele und einige Resultate vorgestellt, die im Rahmen des Forschungsprojekts zur Systementwicklung erzielt worden sind.

Many traditional approaches to reasoning about action and change can be characterised by placing states in a preferential ordering. The potential states after performing an action are predicted to be those satisfying the postconditions of the action that are minimal in this ordering. This is often referred to as preferential entailment and gives rise to a number of solutions to the frame and ramification problems.

Recent literature in this area has seen the appearance of a number of logic-based approaches that employ an explicit notion of causality. In a number of these, the role of causality has been promoted as necessary for solving the frame and ramification problems as well as permitting concise solutions to these problems. However, this has only been demonstrated through representative examples. We attempt to investigate what, if anything, an explicit notion of causality provides. We focus in particular on the causal theory due to McCain and Turner (IJCAI'1995).

Allgemein versteht man unter dem Begriff "Wissensfindung in Daten" (Data Mining, Knowledge Discovery in Databases) den Prozess, relevantes Wissen aus Daten herauszufiltern und in verständlicher Form zu präsentieren. Der Aufdeckung regelhafter Zusammenhänge, die sich gut auf andere Situationen übertragen lassen, kommt dabei eine besondere Bedeutung zu. Das Gebiet wird dominiert von Verfahren, die statistische Methoden mit passenden Heuristiken kombinieren, um den intensionalen Begriff der Relevanz mit objektiver Messbarkeit zu unterlegen.

In diesem Vortrag wird eine andere Perspektive auf diese Problemstellung vorgestellt, die den Prozess der Wissensfindung im Rahmen von induktiver Wissensrepräsentation und nichtmonotoner Inferenz betrachtet. Das ermöglicht nicht nur eine bessere theoretische Fundierung von Methoden, sondern bezieht auch explizit die Verwendung des gefundenen Wissens als Rohstoff für Wissensbasen mit in den Findungsprozess ein. Damit lässt sich die Relevanz von Wissen vor dem Hintergrund geeigneter Formalismen beurteilen. Regeln werden dabei nicht isoliert aufgespürt, sondern im Zusammenspiel mit anderen Regeln betrachtet, wobei Interaktionen zwischen Regeln berücksichtigt und analysiert werden können. Am Beispiel eines probabilistischen Inferenzformalismus wird gezeigt, wie sich mit seiner Hilfe Mengen informativer Regeln aus Häufigkeitsverteilungen berechnen lassen. Ähnliche Verfahren können zur Bestimmung relevanter Default-Regeln aus Repräsentationen epistemischer Zustände angewandt werden.

In this talk we see deductive systems for classical propositional and predicate logic which use deep inference, i.e. inference rules apply arbitrarily deep inside formulas, and a certain symmetry, which provides an involution on derivations. Like sequent systems, they have a cut rule which is admissible. Unlike sequent systems, they enjoy various new interesting properties. Not only the identity axiom, but also cut, weakening and even contraction are reducible to atomic form. This leads to inference rules that are local, meaning that the effort of applying them is bounded, and finitary, meaning that, given a conclusion, there is only a finite number of premises to choose from. The systems also enjoy new normal forms for derivations and, in the propositional case, a cut elimination procedure that is drastically simpler than the ones for sequent systems.

Meta data management problems are pervasive in the development and maintenance of complex application artifacts, such as application programs, databases, and formatted messages. Despite the similarity of solutions to these problems, today they are solved in an application-specific way and usually require much object-at-a-time programming. To make solutions more generic and easier to program, we propose a higher level interface, called Model Management. The main abstractions are models and mappings between models. It treats these abstractions as bulk objects and offers such operators as Match, Merge, Diff, Compose, Extract, and ModelGen. We will present an overview of Model Management followed by some recent research about several of the operators.

Biography:

Philip Bernstein is a Senior Researcher at Microsoft Corporation. Over the past 25 years, he has published over 100 articles on the theory and implementation of database systems, and two books on transaction processing. He is an ACM Fellow, a winner of the SIGMOD Innovations Award, and a member of the National Academy of Engineering.