02 - La fusion d'informations imprécises

Abstract

Possibility theory offers a natural setting for representing imprecise data and poor
\ninformation. This theory turns out to be quite useful for the purpose of pooling
\npieces of information stemming from several sources (for instance, several experts,
\nsensors, or databases) . Indeed it looks more flexible than probability theory for
\nthe representation of aggregation modes that do not express averaging processes .
\nThis paper tentatively explains why possibility theory is appealing for the fusion
\nof imprecise data, and it describes several aggregation modes it allows, along
\nwith their underlying assumptions . The existence of adaptive combination rules
\nare pointed out, that take into account the level of conflict between the sources .
\nThis approach sounds natural in the pooling of expert opinions . It is suggested
\nhere that, under some assumptions, it might also be useful in sensor data fusion .