Adams' high probability deduction and combination of information in the context of product probability conditional event algebra

摘要

Many models of data fusion can be considered to be in "probability-functioal" form -- i.e., as a numerical function of contributing probabilities,with values in the unit interval. such models include: regression models, inference rules, pooling of expert probabilistic opinion, and fuzzy logic transforms of natural language descriptions via the one-point coverages of appropriately chosen random sets. "Conditional event algebra, geared to the analysis of inference rule problems-- and more genrally--"relational event algebra"-- are two new tools for representing such probability-functional models by single -- possibly ocmpounded -- events, leading to a ddper analysis of the joint behavior of the models, e.g., as opposed to comparing the models numerically. This paper has three aspects. First, conditional event algebra is used to provide new insights into deduction problems, emphasizing Adams' "high probability deduction" Second, the concept of "constant-probability events" is used to address the problemof averaging of events. Finally, a new class of relational events is derived.