Graphon Neural Networks and the Transferability of Graph Neural Networks
summary
Constructs a graphon neural network as the limit of GNNs sampled from the same graphon and proves output difference bounds that split into transferable/non-transferable energy terms. The bound depends on graphon smoothness, input L2 sampling error, and a tunable tau bandwidth parameter that trades off spectral discriminability versus nontransferable error. Result: errors vanish as graphs grow under well-separated spectral features; sparse-graph behavior remains open (see uploader link).
AI-written summary · about
chapters (9)
- 0:00Intro
- 0:37Roadmap
- 1:34Network data
- 3:02Graph convolutions and spectral features
- 5:28Graph homomorphisms
- 6:14Graphons
- 7:28Graphon convolutions
- 9:44Transferability of GNNS
- 14:26Conclusion
related talks
01h 28mBoot camp on graphons for graph learning
01h 01mToward Universal Graph Representations: Foundations and Frontiers of Graph Foundation Models
01h 31mGraphons and Graph Limits 2
39mA Local Graph Limits Perspective on Sampling-Based GNNs
16mPreview: Graph NNs
54mGraph Machine Learning & Random Graph Models
39mUsing Graph Neural Networks to Embrace the Dependency Within Your Data
18mTC-GNN: Bridging Sparse GNN Computation and Dense Tensor Cores on GPUs