Workshop paper

[orausch]

The easiest path in is likely some applied workshop/conference.

A good narrative would be to:

1. Find a domain where GraphRNN fails (can be super specific)
2. Tweak the model towards that domain

Alternatively, we could find a metric (rather than a domain) where GraphRNN performs poorly.

[kyduff]

Domains of interest:

1. Social sciences (community/economy/sociological graphs) (sociology, anthropology, economics (trade graphs, globalization metrics), politics (polarization))
2. Medicine & biosciences & neuroscience
3. Logistics & road networks
4. Systems design and component diagrams
5. ASTs & compilers (LLVM IR DFG e.g.)
6. Grid resilience & robustness

[orausch]

On metrics for grid resilience: https://gridarchitecture.pnnl.gov/media/advanced/Electric_Grid_Resilience_and_Reliability.pdf

[orausch]

New Problem Setting:
Given nodes (e.g. Cities) solve a task that connects the nodes (e.g. designing road networks)
Incorporate quality metrics (resilience, etc.) into the loss and get the RNN to "design your road network"