Force aggregation via Bayesian nodal analysis

We describe an adaptive evidence network algorithm developed for a force aggregation, military unit classification tool. The algorithm adaptively selects a partition of the military units of a force based upon probabilistic evidence derived from evolving all-source intelligence data. The partition of the force is a mutually exclusive and exhaustive set of units that accounts for all assets of the military force. The evidence network is isomorphic to the hierarchy of the military force where the units are the nodes of the network, superordinate/subordinate relationships between units instantiate negative evidence arcs, and peer relationships instantiate positive evidence arcs. A genetic algorithm is used to compute the propagation weights of the evidence to ensure that network response is unbiased. The posterior probability for each node is computed using Bayesian methods and is propagated through the network as positive or negative evidence for the inclusion of each unit in the partition. The units of the partition are then used to create a set of unit templates, structures that specify the classification, composition, and command relationships of the military units. The set of templates is subsequently submitted to the cluster tracker and Bayesian classifier of the force aggregation application to perform unit search, track, and classification