Graph_gen.md

Graph Generation Application

Application

Two main methods:

• One-shot generation
• Sequential generation

GraphVAE(One-shot generation)

• Encoder: GNN graph to vector

• Decoder(generator): MLP vector to A E F (graph)

Graph Translation

GT-GAN

GT-GANs learn a conditional generative model, which is a graph translator that is conditioned on input graph and generate the associated target graph.

Aim at translating a graph with one modality to a new one with other modality using deep neural networks architecture. Eg: Examples include generating the traffic jam situation given a road network without traffic jam.

• Graph translator
  • Encoder + decoder
• Conditional graph discriminator

Benchmark dataset

GraphGT

Future Opportunities

• Scalability.
• Validity constraint
• Interpretability and Controllability.