Exploiting Correlations For Efficient Content-Based Sensor Search

Billions of sensor (e. g., in mobile phones or tablet pcs) will be connected to a future Internet of Things (IoT), offering online access to the current state of the real world. A fundamental service in the IoT is search for places and objects with a certain state (e. g., empty parking spots or quiet restaurants). We address the underlying problem of efficient search for sensors reading a given current state - exploiting the fact that the output of many sensors is highly correlated. We learn the correlation structure from past sensor data and model it as a Bayesian Network (BN). The BN allows to estimate the probability that a sensor currently outputs the sought state without knowing its current output. We show that this approach can substantially reduce remote sensor readouts.