A Bayesian Packet Sharing Approach for Noisy IoT Scenarios

Cloud computing and Internet of Things (IoT) represent two different technologies that are massively being adopted in our daily life, playing a fundamental role in the future Internet. One important challenge that need to be handled is the enormous amount of data generated by sensing devices, that make the control of sending useless data very important. In order to face with this challenge, there is a increasing interest about predictive approaches to avoid to send high spatio-temporal correlated data. Belief Propagation (BP) algorithm is a method of performing approximate inference on arbitrary graphical models that is becoming increasingly popular in the context of IoT. By exploiting BP, we can derive effective methods to drastically reduce the number of transmitted messages, while keeping high the data throughput in the global information system. In this paper, we propose a BP approach in a hierarchical architecture with simple nodes, gateways and data centers. We evaluate the error bounding and propose a corrective mechanism to keep a certain quality of the global information in the architecture considered.