Research map / Graph learning / GNN architectures
SizeShiftReg: a Regularization Method for Improving Size-Generalization in Graph Neural Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add SizeShiftReg to standard GNNs | PNA, GCN, and GIN · NCI1, NCI109, PROTEINS, and DD largest 10% | MCC: +22%; +30%; +21% average improvement for PNA, GCN, GIN, respectively | Helped |
| Add SizeShiftReg to PNA | PNA · Deezer, largest 10% | MCC: 0.59 ± 0.06 -> 0.64 ± 0.07 | Helped |
| Add SizeShiftReg to GCN | GCN · Deezer, largest 10% | MCC: 0.49 ± 0.10 -> 0.59 ± 0.06 | Helped |
| Add SizeShiftReg to GIN | GIN · Deezer, largest 10% | MCC: 0.55 ± 0.08 -> 0.61 ± 0.07 | Helped |
| Use all coarsening ratios from 0.1 to 0.9 instead of {0.8, 0.9} (ablation) | PNA + SizeShiftReg · NCI1, NCI109, PROTEINS, and DD | MCC: NCI1: 0.22 ± 0.07 -> 0.18 ± 0.09; NCI109: 0.24 ± 0.07 -> 0.18 ± 0.09; PROTEINS: 0.33 ± 0.09 -> 0.17 ± 0.14; DD: 0.27 ± 0 | Hurt |
| Use very low coarsening ratios (ablation) | PNA + SizeShiftReg · NCI1, NCI109, PROTEINS, and DD | MCC: NCI1: 0.24 ± 0.06 at 0.8 vs 0.07 ± 0.11 at 0.1; NCI109: 0.23 ± 0.07 at 0.8 vs 0.19 ± 0.08 at 0.1; PROTEINS: 0.32 ± 0.11 | Hurt |
| Replace SGC with another coarsening method (ablation) | SizeShiftReg with PNA · NCI1, NCI109, PROTEINS, and DD | MCC | Mixed |
| Replace SGC with spectral clustering (ablation) | SizeShiftReg with GCN · NCI1, NCI109, PROTEINS, and DD | MCC: NCI1: 0.25 ± 0.06 -> 0.28 ± 0.07; NCI109: 0.19 ± 0.06 -> 0.18 ± 0.06; PROTEINS: 0.29 ± 0.13 -> 0.24 ± 0.16; DD: 0.26 ± 0 | Mixed |
| Replace SGC with K-means clustering (ablation) | SizeShiftReg with GIN · NCI1, NCI109, PROTEINS, and DD | MCC: NCI1: 0.23 ± 0.08 -> 0.21 ± 0.09; NCI109: 0.20 ± 0.05 -> 0.20 ± 0.06; PROTEINS: 0.36 ± 0.11 -> 0.35 ± 0.08; DD: 0.25 ± 0 | Mixed |
Closest papers by meaning
- Enhancing Size Generalization in Graph Neural Networks through Disentangled Representation Learning (ICML 2024 · similarity 0.91)
- A Manifold Perspective on the Statistical Generalization of Graph Neural Networks (ICML 2025 · similarity 0.90)
- Towards Bridging Generalization and Expressivity of Graph Neural Networks (ICLR 2025 · similarity 0.90)
- Generalization of Graph Neural Networks Is Robust to Model Mismatch (AAAI 2025 · similarity 0.90)
- Generative Graph Pattern Machine (NeurIPS 2025 · similarity 0.90)
- Graph Classification via Reference Distribution Learning: Theory and Practice (NeurIPS 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.