Utilising energy function and variational inference training for learning a graph neural network architecture.