Fusion of Modular Bayesian Networks for Context-Aware Decision Making

Abstract

Ubiquitous computing brings various information and knowledge derived from different sources, under which Bayesian networks are widely used to cope with the uncertainty and imprecision. In this paper, we propose a modular Bayesian network system to extract context information by cooperative inference of multiple modules, which guarantees reliable inference compared to the monolithic Bayesian network without losing its strength like the ease of management of knowledge and scalability. Moreover, to provide a lightweight updating method for highly complicated environment, we propose a novel method of preserving inter-module dependencies by linking modules virtually, which extends d-separation to an inter-modular concept to control local information to be delivered only to relevant modules. Experimental results show that the proposed modular Bayesian networkscan keep inter-modular causalities in a time-saving manner. This paper implies that a context-aware system can be easily developed by exploiting Bayesian network fractions independently designed or learned in many domains.