A maximum agreement approach to information fusion

Abstract

Information fusion is a generic technique for the problem of information fusion. The common features with this problem are (1) there is a target system X and its state (S) is to be inferred or predicted, (2) there is a group of sensors which have a varying degree of imprecise connection with S of X, and (3) there is a need to come up with an agreed inference or prediction on the state of X (note that consensus here does not mean all with the same opinion by the group). Elsewhere, we proposed an approach called "maximum agreement (MA)" to information fusion in general and probability distribution function aggregation in specific. The basic idea of MA is that an agreed inference is a function of individual sensors' measurements and the agreed measurement can be determined based on the goal that the agreed judgment has a maximum consensus with all individual sensors' measurements. In this paper, we show some alternative methods of MA and discuss their characteristics with reference to MA. We shall then conclude that MA is the best method among all the alternative methods for the problem of expert opinion aggregation or consensus aggregation.