index.md

Graph ML/GNN Reading Group

Meeting Time and Location

Tuesdays 4:30- 6pm at Stanford Gates 415

Introductory Material

• Jure’s CS 224m (Lecture 6 onwards)
• Graph Representation Learning, book by Will Hamilton
• GNN Intro Blog Posts
  • https://distill.pub/2021/gnn-intro/
  • https://distill.pub/2021/understanding-gnns/
• Moses’s repository of survey papers

Schedule

• 4/5/2022: Tutorial on GNN and expressive power

  • Presenter: Weihua Hu
  • Slides
  • Related papers:
    • How Powerful are Graph Neural Networks?

• 4/12/2022: Discussion on expressive power of GNN

  • Presenter: Qian
  • Slides
  • Related papers:
    • How Powerful are Graph Neural Networks? (last week Weihua’s presentation)
    • What Can Neural Networks Reason About
    • Graph Neural Networks are Dynamic Programmers
    • The Exact Class of Graph Functions Generated by Graph Neural Networks

• 4/19/2022: Discussion on expressive power of GNN and open problem

  • Presenter: Amin Saberi

• 4/26/2022: Discussion on expressive power of GNN and open problem

  • Presenter: Yeganeh Alimohammadi
  • Related papers:
    • Generalization and Representational Limits of Graph Neural Networks
    • Topological Graph Neural Networks

• 5/3/2022: Constraint satisfaction and Combinatorial problems

  • Presenter: Aidan Perreault, Moses Charikar
  • Graph Neural Networks for Maximum Constraint Satisfaction
  • Learning a SAT Solver from Single-Bit Supervision
  • Combinatorial optimization and reasoning with graph neural networks

• 5/11: Auto-scaling GNNs & PyG 2.0

  • Presenter: Matthias Fey
  • Slides

• 5/24: GNN Embedding

  • Presenter: Rex Ying

Other Topics and Papers for Future

• Graph generative models: Amin, Yeganeh, Jared, Moses

  • Evaluation Metrics for Graph Generative Models: Problems, Pitfalls, and Practical Solutions (ICLR 22)
  • On Evaluation Metrics for Graph Generative Models (ICLR 22)

• GNN Oversquashing

• On the Bottleneck of Graph Neural Networks and its Practical Implications

• Understanding over-squashing and bottlenecks on graphs via curvature (ICLR 22)

• GNN + geometric representation learning (non-euclidean space)

• Hyperbolic Graph Neural Networks

• Hyperbolic Graph Convolutional Neural Networks

• Geometric Deep Learning, Invariant/equivariant Graph Networks

  • https://www.youtube.com/watch?v=7pRIjJ_u2_c
  • https://www.dropbox.com/sh/u9o69w4rm0uvmzi/AADLGwfdcY94HVsz2eSoCAFPa/invariant%20graph%20networks?dl=0&subfolder_nav_tracking=1
  • E(n) Equivariant Graph Neural Networks

• Simplification

  • Simplifying Graph Convolutional Networks
  • Combining Label Propagation and Simple Models Out-performs Graph Neural Networks

• Hypergraphs

  • https://www.dropbox.com/sh/u9o69w4rm0uvmzi/AADtWutKnbOw0NuxPRpYOwlla/hypergraphs?dl=0&subfolder_nav_tracking=1

• Others

  • Discovering Symbolic Models from Deep Learning with Inductive Biases
  • Applications on knowledge graph, drug discovery etc

Organizers and Participants

• Moses Charikar [email protected]
• Amin Saberi [email protected]
• Jure Leskovec [email protected]
• Qian Huang [email protected]
• Participants