Automatic Distributed Deep Learning Using Resource-Constrained Edge Devices

Processing data generated at high volume and speed from the Internet of Things, smart cities, domotic, intelligent surveillance, e-healthcare systems require efficient processing analytics services Edge to reduce latency response time applications. The fog computing edge infrastructure consists devices with limited computing, memory, bandwidth resources, which challenge construction predictive solutions that resource-intensive tasks for training machine learning models. In this work, we focus on development urban traffic. Our solution is based deep techniques localized in Edge, where have very computational resources. We present an innovative method efficiently gated recurrent-units (GRUs) across available resource-constrained CPU GPU devices. employs distributed GRU model dynamically stops process utilize low-power while ensuring good estimation accuracy effectively. proposed was extensively evaluated using low-powered ARM-based devices, including Raspberry Pi v3 GPU-enabled device NVIDIA Jetson Nano, also compared them Single-CPU Intel Xeon machines. For evaluation experiments, used real-world Floating Car Data. experiments show delivers excellent prediction performance when baseline methods.