convolutional neural networks (CNNs) have achieved great success on grid-like
data such as images, but face tremendous challenges in learning from more
generic data such as graphs. In CNNs, the trainable local filters enable the
automatic extraction of high-level features. The computat