Autoencoder-Aided-Graph-Convolutional-Neural-Network
A novel architecture and training strategy for graph neural networks (GNN). The proposed architecture, named as Autoencoder-Aided GNN (AA-GNN), compresses the convolutional features at multiple hidden layers, hinging on a novel end-to-end training procedure that learns different graph representations per each layer. As a result, the computational scalability improves and the best graph representations at each layer are learnt in a totally data-driven fashion.
