A service oriented Bayesian network architecture

Bayesian network is an effective tool to represent the probabilistic relationships among variables and do probabilistic inference. The algorithms of both network training and reasoning are complex and diverse. There exists no algorithm appropriate to all kinds of dataset, so it is improper to deploy all the algorithms of Bayesian network on one server especially when providing service in a distributed environment. In this paper we integrate Bayesian network into service oriented architecture to propose a loosely coupled Bayesian service architecture, which allows not only a lighter way to adjust the training and reasoning algorithms but also a capability of Bayesian network service provider to deploy the most appropriate service according to actual needs. We then construct a model under the architecture which provides the poor computational capabilities devices with the Bayesian service and offers them the ability to perform a real-time reasoning.