Optimizing dynamic composition of Bayesian Networks for context sensing and inference

Breaking Bayesian Networks for Context Inference from Sensor Networks into smaller Bayeslets is a proven approach for optimizing performance in adaptive resource-constraint ubiquitous computing and networking environments. Automatic selection and composition of such Bayeslets faces the challenge that the related cost factors (inference time, memory consumption) grow exponentially with the number of components. The paper discusses optimising approaches to evaluate the added value of using a particular Bayeslet vs. its cost to prune the dynamic composition graph.