Calculs à l'aide de mots : vers un emploi de termes linguistiques de bout en bout dans la chaîne du raisonnement - TEL - Thèses en ligne Accéder directement au contenu
Hdr Année : 2011

Calculs à l'aide de mots : vers un emploi de termes linguistiques de bout en bout dans la chaîne du raisonnement

Isis Truck

Résumé

Computing with Words, although directly issued from the fuzzy set theory that is now more than 40 years old, is still a widely open research domain. Indeed, one of the dreams of the human beings for a very long time is to create really smart machines. The big passion for this research axis has gone down since the failure of Artificial Intelligence (AI) in the seventies or eighties, notably shown by Hubert Dreyfus’ book published in 1972 and entitled What Computers Can’t Do, and we now know that the robot as an “exact copy of the human being” won’t be happening in a hurry. But, to help humans in their decision making and to simulate their reasoning selectively in some specific and well-defined contexts, is a field that has proven its feasibility and efficiency, and has thus aroused again the interest. The fuzzy set theory then the CW with its various representations of linguistic data, has widely contributed to relaunch this part of AI. In this report, we see how the CW can get involved in this field at different levels. In a practical point of view, many applications had and still can have the benefit of linguistic approaches. Let us cite for the record — and to cite only what we personally tackled — colorimetrics, performing arts and autonomic computing fields into which we have experimented and obtained satisfactory results already. In a theoretical point of view, we consider two axes: – the one that consists in formalizing (to generalize and so, to reuse) new concepts from the introduction of linguistic approaches in applications (for instance the formalization of the LCP-nets), i.e. the axis that feeds on applications, – and the one that consists in setting up existing theoretical concepts better (for instance, to imagine a generalized modus ponens, implications, t-norms, t-conorms, etc. dedicated to linguistic 2-tuples) or to imagine new ones (for instance, to imagine a unification of several linguistic models via a vectorial representation), i.e. the axis that feeds applications. So we show the articulation between theory and practice around the linguistic approaches for the modeling and reasoning under uncertainty, one and the other feeding on each other. From the perceptual computing point of view, we have shown that it can benefit from the four following axes that remain objectives at least partially: – the setting of end-to-end linguistic processes, – the use of various representation models — such as the fuzzy subsets, the linguistic 2-tuples from Herrera & Martínez, proportional 2-tuples from Wang & Hao or linguistic degrees from Truck & Akdag — that allow us to catch the nuances and the results of computation using linguistic definition domains whose cardinality is bounded and close to the user perception, – the setting of linguistic tools to go beyond the well-known fuzzy inference, see for example the LCP-nets with their way to deal with conditional preferences, and – the integration of these tools inside knowledge elicitation processes and the handling of these processes to compose them in order to permit more complex reasoning processes, such as the joint inference in the case of collaborative decision-making.
Le contexte général de mes travaux de recherche se situe dans le domaine de la représentation, modélisation puis combinaison des connaissances en milieu imprécis et vague, en particulier dans la théorie des sous-ensembles flous, d'une part, et dans la théorie des multi-ensembles, d'autre part. Dans ce contexte, nous nous attachons à concevoir des modèles ou outils linguistiques pour palier les problèmes d'imprécision. Notre travail s'est d'abord fondé sur la modulation, préservant la simplicité d'un espace de variables linguistiques de cardinalité assez faible mais sur lequel un large éventail de nuances peut être appliqué. Puis nous avons utilisé ces modèles et outils dans des contextes où l'expression de nuances est très importante, comme le *perceptual computing* dans le domaine de la classification des couleurs ou encore le jeu théâtral, et, plus récemment, pour la capture des intentions des programmeurs pour établir des politiques de contrôle et d'adaptation dynamique des architectures logicielles selon l'approche du calcul auto-régulé (*autonomic computing*).
Fichier principal
Vignette du fichier
HDR.pdf (8.51 Mo) Télécharger le fichier

Dates et versions

tel-00639737 , version 1 (10-11-2011)

Identifiants

  • HAL Id : tel-00639737 , version 1

Citer

Isis Truck. Calculs à l'aide de mots : vers un emploi de termes linguistiques de bout en bout dans la chaîne du raisonnement. Intelligence artificielle [cs.AI]. Université Paris VIII Vincennes-Saint Denis, 2011. ⟨tel-00639737⟩
199 Consultations
101 Téléchargements

Partager

Gmail Facebook X LinkedIn More