mlsanityData to 7 Oct 2026

Train Small, Deploy Large: Zero-Shot GNN Transfer Through Geometric Renormalization

arXiv 2607.27767 ↗First posted 30 Jul 2026cs.LG cs.AI physics.soc-phcited by 0

Not on the map. Why?

Abstract (arXiv)

Graph neural networks (GNNs) can operate on large graphs but become infrastructure-sensitive at the scale of millions of nodes and typically require scalable training techniques for even larger graphs. This raises a central question: when can a model trained on a smaller, scaled-down replica of a graph be deployed on the full-resolution graph without retraining? We introduce a zero-shot transfer protocol in which a GNN is trained on a graph coarse-grained by geometric renormalization (GR), and the resulting weights are transferred directly to the original network. Across synthetic and real-world networks, training on GR scaled-down replicas preserves much of the original-scale predictive performance while significantly reducing training cost. We further find that learned representations and predictive trajectories remain aligned across scales. These findings suggest that structural similarity may be more important than network size in determining GNN transferability, opening a path toward scale-equivariant graph architectures.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 32 of its 53 references that are in the map. First posted on arXiv 30 Jul 2026.

Builds on

32 of 53 references.

References · 32

  1. 2016Semi-Supervised Classification with Graph Convolutional Networkscited by 11,282
  2. 2017Graph Attention Networkscited by 8,854
  3. 2017Inductive Representation Learning on Large Graphscited by 6,284
  4. 2019Fast Graph Representation Learning with PyTorch Geometriccited by 2,414
  5. 2019Similarity of Neural Network Representations Revisitedcited by 1,695
  6. 2018Pitfalls of Graph Neural Network Evaluationcited by 994
  7. 2016Revisiting Semi-Supervised Learning with Graph Embeddingscited by 970
  8. 2018Hierarchical Graph Representation Learning with Differentiable Poolingcited by 786
  9. 2019GraphSAINT: Graph Sampling Based Inductive Learning Methodcited by 618
  10. 2019Graph U-Netscited by 489
  11. 2020Wiki-CS: A Wikipedia-Based Benchmark for Graph Neural Networkscited by 253
  12. 2020From Local Structures to Size Generalization in Graph Neural Networkscited by 119
20 more
  1. 2020Graphon Neural Networks and the Transferability of Graph Neural Networkscited by 117
  2. 2021Generalized Shape Metrics on Neural Representationscited by 109
  3. 2021Scaling Up Graph Neural Networks Via Graph Coarseningcited by 80
  4. 2021Understanding Pooling in Graph Neural Networkscited by 69
  5. 2021Transferability Properties of Graph Neural Networkscited by 42
  6. 2020Network Geometrycited by 31
  7. 2016Machine learning meets network science: dimensionality reduction for fast and efficient embedding of networks in the hyperbolic spacecited by 21
  8. 2021Graph coarsening: From scientific computing to machine learningcited by 21
  9. 2019Mercator: uncovering faithful hyperbolic embeddings of complex networkscited by 10
  10. 2023A Metadata-Driven Approach to Understand Graph Neural Networkscited by 10
  11. 2024Graph Pooling via Ricci Flowcited by 10
  12. 2024Graph Coarsening with Message-Passing Guaranteescited by 9
  13. 2018Scale-free Networks Well Donecited by 7
  14. 2021Detecting the ultra low dimensionality of real networkscited by 5
  15. 2025Geometry-Aware Edge Pooling for Graph Neural Networkscited by 5
  16. 2019Geometric renormalization unravels self-similarity of the multiscale human connectomecited by 4
  17. 2024Feature-aware ultra-low dimensional reduction of real networkscited by 3
  18. 2025Performance Heterogeneity in Graph Neural Networks: Lessons for Architecture Design and Preprocessingcited by 2
  19. 2025Chordless cycle filtrations for dimensionality detection in complex networks via topological data analysiscited by 1
  20. 2026Graph Neural Networks Are Not Continuous Across Graph Resolutionscited by 1

Built on by

Not cited yet.