Classification for People Distribution with Reservoir Computing <Abstract> The wireless sensor network (WSN) that collects information from sensor nodes installed in the environment using wireless communication can achieve automatic and continuous data collection. The framework of data collection and information processing by sensor networks is called the Internet of Things (IoT), and much research and industrial applications are attracting attention as one of the important technologies in recent years. The huge amount of data continuously collected by WSN causes a problem of consuming enormous power to process information in the data server. In addition, the amount of data that can be transmitted from the sensor node is reaching its limit due to the limitation of the wireless communication capacity in the network. In order to solve these problems, distributed information processing techniques such as edge computing has been proposed as a technology to reduce the power consumption and the amount of data transmitted over the network in information processing on data servers. By the way, the human brain is known to perform complex information processing with low power consumption. Spiking Neural Network (SNN) consists of spiking neurons, which communicate information one another by spikes, imitates biological experimental-observed dynamics of the neuron. In order to develop new intelligent information processing, there have been many studies on learning and hardware implementation using SNN. The concept of Wireless Brain-Inspired Computing(WiBIC) was proposed. In the WiBIC concept, the neurons are linked by wireless communications. Furthermore, the WSN acquires information processing functions by adding neuron functions to sensor nodes. In other words, WiBIC can be regarded as a technology for distributed information processing. As a first step for realizing the WiBIC concept, the wireless SNN system was proposed and implemented. In this research, we propose the learning model for environmental sensor data under the SNN framework. The proposed learning model achieved the classification of the number of people into three classes with an accuracy of $0.81$. |