Hierachical Graph Convolutional Neural Network able to integrate data from multiple graphs using embedding and projection operations in a residual approach. Based on spatial graph convolutions.
Applied to Census data to predict election outcomes - provides significant improvement in RMSE - a 3 fold decrease compared to generalised linear methods.
Results to be published in AAAI 2019 conference output.