Querying graphs and their representations

Table of Contents

1. About The Project
2. Fall Summary
3. Winter Summary
4. Spring Summary
5. Technology
6. License
7. Contact

About The Project

• Graph neural networks transform the nodes of a graph into a high dimensional latent space.

• This project will contrast the distances between nodes of a graph in the input space (graph structure) to their embedding in the latent space.

• Queries of interest will be finding node/subgraph outliers, and comparing representations produced by different deep learning methods.

• Project can be extended to consider different ways of reducing distortions in embeddings and measuring the local dimensionality of the embedding space.

Fall Summary

Research Proposal

Proposal Slides

Winter Summary

Research Log

Spring Summary

Research Log

Technology

As we are investigating two different database softwares to build on top of, we have two different forks of those repositories.

Deprecated:

• ArangoDB fork: arangodb_ersp
• OpenCypher fork: openCypher_ersp

Active:

• Minimum Edge Set Perturbation: minimum-edge-set-perturbation
  • Homophily: homophily
  • Degree: degree
  • Common Neighbors: common-neighbors

License

Distributed under the MIT License. See LICENSE for more information.

Contact

Will Corcoran - wcorcoran@ucsb.edu

Wyatt Hamabe - whamabe@ucsb.edu

Niyati Mummidivarapu - niyati@ucsb.edu

Danish Ebadulla - danish_ebadulla (at) umail (dot) ucsb (dot) edu