LIP6 1999/004:
THÈSE de DOCTORAT de l'UNIVERSITÉ PARIS 6 LIP6 /
LIP6
research reports
218 pages - Décembre/December 1998 -
French document.
PostScript : 698 Ko /Kb
Contact : par mail / e-mail
Thème/Team: Systèmes d'Aide à la Décision et à la Formation
Titre français : Représentation du raisonnement humain dans la décision : Application à la photo-interprétation
Titre anglais : A modelling of human reasoning in decision problems and an application to interpretation in remote sensing activities
Abstract : The limitations encountered by expected utility theory in the modelisation of effective decision making led the path to a model of decision making which takes into account the way preferences are handled according to reasoning. The idea consists in introducing reasoning by similarity as the type of reasoning used in decision problems: The solution to a current problem is searched among maximal solutions (according to the preferences) of similar memorised decision problems. The similarity of two decision problems is defined, according to the current state of reasoning, by seeking partial identities between the descriptions of the two problems while neglecting descriptive features which distinguish the two problems. The partial identity depends upon previously used negligibilities, and is accepted or rejected with respect to the goals and state of the reasoning. The preference relation is defined using tuples which allows the expression of preferences between different partial identities and different negligibilities. Using these two notions, negligibility and similarity, we introduce two relations of entailment by similarity, and two fixed point semantics, the first relation being named strong relation of entailment allows a first formalisation of reasoning by similarity, but requires the completeness of the preference relation (or else the reasoning may not conclude). The second relation of entailment allows a second formalisation of reasoning by similarity without requiring the completeness of the preference relation. The implementation in the Soar system consists in transforming the inference mechanism from a pattern matching process to a partial one, and in taking into account negligible features, the preferences formally defined being expressible in the Soar system. This implementation can take into account problem solving under uncertainty, for example remote sensing image interpretation, by using similarity to handle uncertainty.
Key-words : Decision, Non monotonic logics, Similarity, Soar
Publications internes LIP6 1999 / LIP6 research reports 1999