Graph Neural Networks
Current deep learning tools are typically used on the data with limited form of sequences (like text or speech data ) or grids (like image data).
We want to develop neural networks that is applicable to more kinds of data like graph data besides sequences and grids data.
adopt ideas of representation learning to nodes so that similar nodes in the network have similar embeddings(a d dimensional real vector).
Nodes aggregate information from their neighbors by neural networks. Each node defines a computation graph.
