FedGCN: Federated Learning-Based Graph Convolutional Networks for Non-Euclidean Spatial Data

Federated Learning (FL) can combine multiple clients for training and keep client data local, which is a good way to protect privacy. There are many excellent FL algorithms. However, most of these only process with regular structures, such as images videos. They cannot non-Euclidean spatial data, that is, irregular data. To address this problem, we propose Learning-Based Graph Convolutional Network (FedGCN). First, (GCN) local model FL. Based on the classical graph convolutional neural network, TopK pooling layers full connection added improve feature extraction ability. Furthermore, prevent from losing information, cross-layer fusion used in GCN, giving an ability Second, paper, federated aggregation algorithm based online adjustable attention mechanism proposed. The trainable parameter ρ introduced into mechanism. method assigns corresponding coefficient each model, reduces damage caused by inefficient parameters global improves fault tolerance accuracy algorithm. Finally, conduct experiments six datasets verify proposed not has but also certain degree generality. perform well different networks.