Isabelle Bloch

			TELECOM ParisTech (ENST)
			CNRS UMR 5141 LTCI
			Paris, France
			Isabelle.Bloch@enst.fr
			http://www.tsi.enst.fr/~bloch/
			

Morphologie mathématique floue, applications en raisonnement spatial et en logique

La morphologie mathématique repose sur le cadre algébrique des treillis complets et des adjonctions, lui conférant des propriétés fortes et permettant de multiples extensions. En particulier, les extensions au flou des principales opérations de morphologie mathématiques, telles que la dilatation et l'érosion, peuvent être faites en préservant toutes les propriétés de ces opérateurs. Ces extensions ont de multiples applications. Nous en présenterons deux. La première concerne la définition de relations spatiales, pour des applications en raisonnement spatial et en reconnaissance de structures dans les images guidée par un modèle. La seconde application concerne la logique et nous l'illustrerons sur un exemple en médiation.

La présentation de l'exposé d'Isabelle Bloch : ici

Uzay Kaymak

			Econometric Institute, Erasmus School of Economics 	
			Erasmus University Rotterdam, 
			Rotterdam, the Netherlands
			kaymak@few.eur.nl
			http://people.few.eur.nl/kaymak/
			

Using Linguistic Information in Density Estimation : a case for probabilistic fuzzy systems

Being universal function approximators, fuzzy systems implement a nonlinear input-output mapping in most applications. This mapping is deterministic even when the fuzzy system parameters are determined in a data-driven manner. The data available, however, is almost always noisy and is distributed in a certain way, which seems to be ignored in most fuzzy modelling applications. In other words, probabilistic uncertainty in the data is not considered explicitly in most fuzzy models. In this presentation, we discuss how fuzzy systems could be extended to estimate probabilistic uncertainty in the form of probability density functions. Probabilistic fuzzy systems are proposed for this purpose. Compared to the conventional density estimation techniques, probabilistic fuzzy systems emphasize incorporating expert knowledge in the models and the use of linguistic information to constrain the identified models. In this way, models are brought in line with the linguistic semantics provided by the modeller. We discuss extension of the existing types of fuzzy systems to their probabilistic fuzzy equivalents and the estimation of their parameters. Examples from the financial domain are given to illustrate the use of probabilistic fuzzy models.

La présentation de l'exposé d'Uzay Kaymak : ici

Ronen Brafman

			Department of Computer Science
			Ben-Gurion University 	
			Beer-Sheva, Israel
			brafman@cs.bgu.ac.il
			http://www.cs.bgu.ac.il/~brafman/
			

Working with Preferences

Preference modeling has a long tradition in economics and decision analysis, and the need to elicit, model, and reason with user preferences is now well recognized in AI. In this talk I will discuss some of my own work in this area, which is motivated by the need to provide simple and intuitive languages for modeling preferences that mimic expressions used in natural language. I will describe the CP-nets model and one of its extensions, and explain the role the graphical structure plays in algorithms for computing optimal choices and in their complexity analysis. Time permitting, I will describe applications from the area of adaptive content selection.

La présentation de l'exposé de Ronen Brafman : ici