Distributed Execution of Deep Learning in Wireless Sensor Networks
In wireless sensor networks, microcontrollers associated with sensors are becoming more sophisticated and power efficient. This allows task processing such as learning, anomaly detection, and decision making, which were previously performed in the cloud, to be offloaded to the sensor network, enabling them to be performed efficiently at locations close to where the data is generated. Autonomous intelligent sensor networks can be realized.In this study, we propose a new architecture for distributed execution of CNNs in a local wireless sensor network consisting of sensor devices that are the source of data, and propose a distributed execution protocol and algorithm for this purpose. The proposed method is based on the data (e.g., temperature distribution) acquired periodically and arealistically by a mesh wireless sensor network, and assigns the role of a unit in deep learning to a sensor node.
Y. Fukushima, D. Miura, T. Hamatani, H. Yamaguchi and T. Higashino, "MicroDeep: In-network Deep Learning by Micro-Sensor Coordination for Pervasive Computing," 2018 IEEE International Conference on Smart Computing (SMARTCOMP), 2018, pp. 163-170, doi: 10.1109/SMARTCOMP.2018.00087.