A Graph-Based Vertical Federation Broad Learning System

A broad learning system is a lightweight deep neural network with breadth expansion, which is widely used in face recognition, error detection, and so on. The broad learning system can make full use of grid data, but it is not suitable for utilizing graph data that can represent data relationships. Due to the emphasis on data privacy, the feature data of graph for training models is often in a fragmented state. Based on the above problems, this paper introduces the vertical federation idea and the graph convolutional neural network into the broad learning system and uses the graph neural network to assist the broad learning system in extracting features of graphs. We use the isolated graph information jointly extracted by the vertical federation framework for broad learning and propose a graph-based vertical federation broad learning system. Since the weights for extracting features are randomly generated during the initial graph establishment phase. There is no guidance for extracting features, and the quality of the extracted features is not guaranteed. Therefore, this paper introduces the extreme learning auto-encoder into the graph-based vertical federation broad learning system to generate weights for extracting graph features.