Conférences invitées

Pattern Mining: The Past, Present, and Future

Mohammed Zaki
Rensselaer Polytechnic Institute
Troy, New York (USA).

Abstract: Pattern mining is one of the fundamental techniques in data mining. As one increases the complexity of the pattern types, from subsets, to subsequences, subtrees, and subgraphs, one discovers potentially more informative patterns. In this talk I will offer a tour of the past and the present research landscape in this area, and I'll conclude with some thoughts on directions for the future.

Bio: 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. 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 associate editor for Data Mining and Knowledge Discovery, ACM Transactions on Knowledge Discovery from Data, Statistical Analysis and Data Mining, Knowledge and Information Systems, and International Journal of Knowledge Discovery in Bioinformatics. He was the program co-chair for SDM'08 and SIGKDD'09, and is co-chairing PAKDD'10. He received the National Science Foundation CAREER Award in 2001 and the Department of Energy Early Career Principal Investigator Award in 2002. He also received the ACM Recognition of Service Award in 2003, and an IEEE Certificate of Appreciation in 2005.

Recent Advances in Partitioning Clustering Algorithms for Interval-Valued Data

Francisco de Assis Tenório de Carvalho
Universidade Federal de Pernambuco
Recife, Brésil.

Abstract: Cluster analysis have been widely used in numerous fields including pattern recognition, data mining and image processing. Their aim is to organize a set of objects into clusters such that items within a given cluster have a high degree of similarity, whereas items belonging to different clusters have a high degree of dissimilarity.

In particular, partitioning clustering models aims to organize a set of items into a pre-defined number of clusters. Our reference clustering algorithms are the so-called partitioning dynamic cluster algorithms. They are iterative two steps relocation clustering algorithms involving at each iteration the construction of the clusters and the identification of a suitable representative or prototype (means, factorial axes, probability laws, etc.) of each cluster by locally optimizing an adequacy criterion between the clusters and their corresponding prototypes.

Often, objects to be clustered are represented as a vector of quantitative features. However, the recording of interval-valued data has become a common practice in real world applications and nowadays this kind of data is often used to describe objects. Symbolic Data Analysis (SDA) is an area related to multivariate analysis, data mining and pattern recognition, which has provided suitable data analysis methods for managing objects described as a vector of intervals.

In this talk, we review partitioning clustering algorithms for interval-valued data having as reference the dynamic clustering algorithm. For each clustering algorithm, it is given the clustering criterion, the best ptototype of each cluster, the best distance associated to each cluster (if any) as well as the best partition in a fixed number of clusters. Moreover, various tools for the partition and cluster interpretation of interval-valued data furnished by these algorithms are also presented. Finally, in order to show the usefulness of these algorithms and the merit of the partition and cluster interpretation tools, experiments with real interval-valued data sets are given.

Action Rules and Meta-actions

Zbigniew W. Ras
University of North Carolina
Charlotte, North Carolina (USA).

Abstract: Action rules describe possible transitions of objects in a decision system S from one state to another with respect to the decision attribute. Classification attributes in S are partitioned into stable and flexible. Meta-actions are defined as actions which trigger changes of flexible attributes in S either directly or indirectly because of correlations among certain attributes in the system. In medical area, taking a drug is an example of a meta-action since some test results (values of attributes in S) for a given patient will get changed. Laboratory and radiological tests are examples of classification attributes which are flexible.

Early research on action rule discovery followed rule-based approach and it required extraction of classification rules from a decision system before constructing any action rule. Newest algorithms follow an object-based approach and they extract action rules directly from a decision system. In this presentation we will show how meta-actions can be used to identify interesting action rules and action rules of the lowest cost.

Combiner approche logique et numérique pour la réconciliation de données et l'alignement d'ontologies

Marie-Christine Rousset
Laboratoire d'Informatique de Grenoble (LIG)
Grenoble, France.

Résumé : La réconciliation de données et l'alignement d'ontologies sont deux problèmes au coeur de l'intégration d'informations, auxquels de nombreux travaux se sont attaqués soit par une approche symbolique, soit par une approche numérique.

Dans cet exposé, nous montrons que combiner une approche logique et une approche numérique permet d'une part d'obtenir des mises en correspondance moins empiriques et qui ont plus de sens, et d'autre part un meilleur passage à l'échelle en exploitant au mieux l'inférence logique pour élaguer certains calculs numériques.

Nous nous appuyons sur la logique, les systèmes d'équations ou les probabilités comme des formalismes mathématiques complémentaires permettant de traduire et exploiter la sémantique d'un schéma médiateur ou d'une ontologie.

Pour illustrer notre propos, nous présentons d'abord une méthode de réconciliation de références qui s'applique à des sources de données RDF conformes à un même schéma.

Cette méthode commence par inférer des décisions certaines de réconciliation et de non réconciliation qui découlent logiquement des contraintes sémantiques déclarées au niveau du schéma. Puis elle calcule la similarité entre les paires de références non traitées par la méthode logique par la résolution d'un systèmes d'équations modélisant les dépendances entre réconcilations découlant de la description sémantique associée à chaque référence.

Nous présentons ensuite une méthode d'alignement d'ontologies qui combine inférence logique et classification bayésienne pour estimer la probabilité de mappings entre classes à partir des méta-données des instances déclarées ou inférées dans ces classes. Cette méthode s'appuie sur un algorithme qui, à partir d'un ensemble de mappings candidats qui peut être l'ensemble de tous les mappings possibles, fournit l'ensemble des mappings dont la probabilité dépasse un certain seuil. Cet algorithme passe à l'échelle grâce à l'exploitation d'une propriété de monotonie de la fonction de probabilité par rapport à un ordre sur les mappings fondé sur leur sémantique logique.