CONCEPTUAL STRUCTURES: FROM INFORMATION TO INTELLIGENCE

Invited Speakers


Speaker Professor Michel Chein
Title Entities and Surrogates in Knowledge Representation
Abstract

The question of the relationships between a word, or a text, or a symbol, and the object or concept, or idea to which it refers is a fundamental problem in many domains: philosophy, linguistics, psychology, etc. This reference problem has been tackled in many domains of Computer Science, especially in databases integration (e.g., entity resolution, record linkage, duplicates elimination, reference reconciliation) and computational linguistics (e.g., disambiguation, referring expressions). Most approaches are statistical, e.g., the object identification problem is viewed as a classification problem, but recent works, as ours, use AI techniques. We propose in this talk a simple logical framework for studying the relationships between a surrogate (a symbol in a computer system) and an entity to which it refers in an application domain. The two worlds linked by such a reference relation are irreconcilable ("La rèalitè est impossible" said Jacques Lacan), thus there is no hope to automatically solved reference problems. Nevertheless, if knowledge are used, it can be possible to help users faced with reference problems. In the proposed framework, which is motivated by actual problems in bibliographical databases, knowledge are described in terms of first order logic or in terms of conceptual graphs.
Bio-Data Michel Chein is Professor Emeritus in Computer Science at the University of Montpellier. He founded in 1992, with Marie-Laure Mugnier, the research group on Conceptual Graphs at the Laboratory for Informatics, Robotics and Microelectronics of Montpellier. They both recently published the book "Graph-based Knowledge Representation", Springer 2009. From 1965 to 1970 he was researcher at CNRS (Centre National de la Recherche Scientifique) at Grenoble, from 1970 to 1972 he was Professor of Applied Mathematics at the Univ. of Le Mans, from 1972 to 1980 he was Professor of Computer Sc. at the Univ. Paris 6, since 1980 he is professor of Computer Science at the University of Montpellier. Among various activities, he was the founder and director of the CRIM (Centre de recherches en informatique de Montpellier (1982-85)) and the founder and director of the school for doctoral studies: "Information, Structures, Systèmes", Univ. Montpellier (1994-2000). From 1965 to 1985 his research interests were graph and order theory, since 1985 he mainly works in Artificial Intelligence, and more specifically in Knowledge Representation (for more details see http://www.lirmm.fr/~chein).
Speaker Dr. Boris Motik
Title Combining Description Logics, Description Graphs, and Rules
Abstract

Recent practical experience with description logics has revealed that their expressivity is often insufficient to accurately describe structured objects—objects whose parts are interconnected in arbitrary, rather than tree-like ways. To address this problem, I shall present an extension of DL languages with description graphs—a modeling construct that can accurately describe objects whose parts are connected in arbitrary ways. To enable modeling the conditional aspects of structured objects, the description graph formalism is extended with rules. I shall present examples demonstrating the expressivity of the new formalism and will also discuss the computational properties of the hybrid formalism.
Bio-Data - BSc in Electrical Engineering, University of Zagreb, Croatia, 1996
- MSc in Computer Science, University of Zagreb, Croatia, 1999
- PhD in Computer Science, University of Karlsruhe, Germany, 2002

I am currently a lecturer in computer science at the University of Oxford, UK. My research interests are knowledge representation, ontologies, semistructured data and databases, and the Semantic Web.
Speaker Professor Mohammed Zaki
Title Practical Graph Mining
Abstract

Given the ubiquity of large-scale graphs and networks, graph mining has rapidly grown to occupy a center-stage within data analysis and mining. In this talk I will present our recent work on mining interesting, representative subgraph patterns from very large graph databases. Some aspects of graph indexing may also be covered. I'll conclude with thoughts on future challenges and research directions.
Bio-Data Mohammed J. Zaki is a Professor of Computer Science at RPI. He received his Ph.D. degree in computer science from the University of Rochester in 1998. His research interests focus on developing novel data mining techniques, especially in bioinformatics. He has published over 175 papers and book-chapters on data mining and bioinformatics. We was the founding co-chair for the BIOKDD series of workshops. He is currently an Executive Editor for Statistical Analysis and Data Mining, and an Associate Editor for Data Mining and Knowledge Discovery, ACM Transactions on Knowledge Discovery from Data, Knowledge and Information Systems, ACM Transactions on Intelligent Systems and Technology, Social Networks and Mining, and International Journal of Knowledge Discovery in Bioinformatics. He was the program co-chair for SDM'08, SIGKDD'09 and PAKDD'10. He received the National Science Foundation CAREER Award in 2001 and the Department of Energy Early Career Principal Investigator Award in 2002.
Speaker Professor Jerome Lang
Title Graphical representation of ordinal preferences : languages and applications
Abstract

The specification of a decision making problem includes the agent's preferences on the available alternatives. The choice of a model of preferences (such as utility functions or binary relations) does not say how preferences should be represented (or specified). Writing them explicitly, for instance by ranking all alternatives one after the other, is unrealistic when the set of alternatives is large, and in particular when it has a combinatorial (or multiattribute) structure, i.e., when each alternative consists of a tuple of values, one for each of a given set of variables.

For such combinatorial domains, we need languages allowing to express preferences as succinctly as possible. Such compact preference representation languages have been studied in the Artificial Intelligence research community. A significant number of these languages are 'graphical', because they consist of a graphical component describing preferential dependencies between variables, together with a collection of local preferences on single variables or small subsets of variables, compatible with the dependence structure.
In this talk I will focus on graphical languages for ordinal preferences, and especially on CP-nets and their extensions and variants. After giving an brief presentation of these languages, I will show (mostly using examples) how they can be used for individual, collective or distributed decision making.
Bio-Data Jérôme Lang is a senior researcher ("directeur de recherche'') at Centre National de la Recherche Scientifique. Since 2008 he is affiliated with the Laboratoire d'Analyse et de Modélisation de Systèmes d'Aide à la Décision (LAMSADE), Université Paris-Dauphine. From 1991 to 2008 he was a CNRS researcher at Institut de Recherche en Informatique de Toulouse. His research interests span a large part of Artificial Intelligence, especially Knowledge Representation and Multi-Agent Systems. His recent research activities focus on preference representation and computational social choice.
Speaker Professor Bernhard Ganter
Title Exploring Conceptual Possibilities
Abstract

Shortly after the ideas of Formal Concept Analysis were first presented by Rudolf Wille in his seminal paper of 1982, one of the basic FCA methods came up: Attribute Exploration. It offers an interactive technique for exploring the possible attribute combinations in a given domain, supporting the search by a powerful and still somewhat peculiar algorithm. Since then, remarkable progress has been made, so that the theoretical foundations of Formal Concept Analysis nowadays are broad and wellestablished. There are still remarkable research activities in the field, but many of the basic questions are solved and one may wonder what the future directions of research might be. What are worthwhile directions of further investigations on conceptual structures? Suggestions for answers may be obtained from applications of the attribute exploration method, which, when applied to real world situations, often is confronted with problems that require more than the basic technique. Many extensions of the method, with additional features, have been discussed and investigated, mainly because there was demand from the side of applications. Quite in the beginning it was studied how “background knowledge” can be taken into account. An natural question also is how incomplete and imprecise data can be handled. The study of data with symmetries led to an extensions of the method to predicate logic (Horn formulae). More recently, several attempts were made to handle structured data as well. What if the objects under consideration have an inner structure that is related to their attributes? For example, if the objects are molecules, or mathematical items. Or what, if the objects are related to other objects, as it is the case in processes or in causal networks? Then more expressive logics, like description logics, are needed, but that raises difficult questions. And perhaps even more challenging are situations where the data are unreliable, not merely because they are imprecise or incomplete, but because they are provided by many users not all of whom can be trusted. We give an overview of our present knowledge of this theme and indicate some possible goals.
Bio-Data