Distributed Training of Graph Convolutional Networks

The aim of this work is to develop a fully-distributed algorithmic framework for training graph convolutional networks (GCNs). proposed method able exploit the meaningful relational structure input data, which are collected by set agents that communicate over sparse network topology. After formulating centralized GCN problem, we first show how make inference in distributed scenario where underlying data split among different agents. Then, propose gradient descent procedure solve problem. resulting model distributes computation along three lines: during inference, back-propagation, and optimization. Convergence stationary solutions problem also established under mild conditions. Finally, an optimization criterion design communication topology between order match with describing relationships. A wide numerical results validate our proposal. To best knowledge, combining neural