Distributed Training from Multi-sourced Data

A distributed system is a set of logical or physical units capable of performing calculations and communicating with each other. Nowadays, these systems are at the heart of technologies such as the Internet of Things IoT, the Internet of Vehicles IoV, etc. These systems collect data, perform calculations and make decisions. On the other hand, deep learning (DL) has led to enormous progress in the field of artificial intelligence. Since the precision of DL to form a set of reference data on a single machine is known, it becomes more interesting to form several models and distribute the intelligence over the different nodes of the system by different calculation strategies.

In this paper we propose a method using deep neural networks on several machines by distributing the dataset before starting the training, ensuring communication between them in order to improve the calculation time and accuracy.