The exploitation of Artificial Intelligence algorithms in real problems is a very interesting method for their improvement. Actually, these applications have such constraints that any weakness of the intended algorithm is exposed, far quicker than using the usual academic problems. More precisely, in my PhD thesis, I studied two medical diagnosis helping projects. Thus, each tool I have created is constantly interacting with a medical team. One of my contributions consists in associating the reasoning capabilities of Markov models with intuitive-looking fuzzy sets. The resulting system has strong diagnosis capabilities, primarily based on this interaction. In particular, I propose a new method for learning directly from a diagnosis. Its application is useful when a doctor needs modifying the system's diagnosis. At this moment, the system starts learning from this correction to enhance the patient profile, so that the new diagnosis is compatible with the doctor's one while ensuring some numerical stability properties. Using this method, any doctor is able to correct a patient's profile without manually modifying every parameter of the model. Finally, I explain we can generalise this approach to apply it to non-medical applications. This is exposed in a classical problem, the localisation of a mobile robot that evolves in a structured environment. This generalisation allowed the realisation of a generic Object Oriented package. With this software, a given user could create a whole application, merely by putting aside some modules and linking them altogether so that the required computations are done in the right order. Thanks to its Object nature, this package allows for easily integrating and updating modules along their evolution. Finally, such an interface should help the creation of new applications by reducing the amount of time that is necessary. Actually, once the user has chosen the modules, their order and their parameters, the application handles the whole code generation to obtain a complete module.

EXPLOR_STEP=$WICRI_ROOT/Wicri/Lorraine/explor/InforLorV4/Data/Hal/Curation

HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 005684 | SxmlIndent | more

HfdSelect -h $EXPLOR_AREA/Data/Hal/Curation/biblio.hfd -nk 005684 | SxmlIndent | more

{{Explor lien
   |wiki=    Wicri/Lorraine
   |area=    InforLorV4
   |flux=    Hal
   |étape=   Curation
   |type=    RBID
   |clé=     Hal:tel-00003378
   |texte=   Apprentissage et adaptation pour la modélisation stochastique de systèmes dynamiques réels
}}