A distributed inference system on sensor nodes using neighbors' context data

Recently, several studies have been made on techniques for the inference of contexts in indoor environments using sensor nodes attached to various objects. In existing systems, a sensor data server accumulates all data received from sensor nodes to infer contexts. However, when the number of nodes increases, the volume of the data concentrated on the server increases, and the calculation cost at the server often explodes. In order to solve this problem, we propose a distributed inference system on sensor nodes that infers the nodes' local contexts. The system has three remarkable features: 1) sensor nodes which can detect neighboring nodes, 2) inference engines on sensor nodes, which infer nodes' local contexts, and 3) sharing context data between neighbor nodes. This paper describes the implementation methods, composition, and operation examples of the system.