Bipolar radial basis function inferencing networks

In this letter we introduce bipolar radial basis function inferencing networks (BRBFNs) where nodes represent concepts and descriptive attributes, and connections approximate generic inter-concept relational strengths in the interval [-1,1]. The inference mechanism in such networks is based on a process which we call Ensemble Coherence Modulation (ECM) in which subsets of concepts are either excited or inhibited depending upon their degree of relational coherence with a specified set of concepts. ECM is emulated by spreading network activity through a bipolar radial basis signal function. Inductive learning in such systems takes place through a heuristic learning law. Simulation results presented demonstrate the efficacy of employing the BRBFN model for a variety of applications such as inferencing with conflicting knowledge, evidence aggregation, knowledge induction, incorporating context in inferences, and for abduction and deduction.