The dynamic of belief in the transferable belief model and specialization-generalization matrices
UAI '92 Proceedings of the eighth conference on Uncertainty in Artificial Intelligence
Hi-index | 0.00 |
After receiving a piece of evidence Π, intelligent agents adapt their belief state. Knowing the belief state of several sensor agents, a fusion agent forms its own belief state. A belief state (old or new) is characterized by numbers μ(A) attached to subsets A of the universe of worlds. (Usually μ(A) is interpreted as that part of the belief of an agent in the sentence "the actual world belongs to A" and not in any sentence "the actual world belongs to a strict subset of A".) In this paper we study how fusion agents treat changes of belief state of sensors. Two kinds of fusion are considered: the averaging by convex combination, and Dempster type conjunctive rules of combination. We establish that there is no coherent treatment, at the fusion agents' level, that is compatible with both kinds of fusion. To deal with better detection, a mixed structure involving intermediate supervision agents is proposed.