Research map / Optimization / Implicit bias and generalization
Fast Learning of Graph Neural Networks with Guaranteed Generalizability: One-hidden-layer Case
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 |
|---|---|---|---|
| Use Heavy Ball momentum instead of vanilla GD | GNN learning algorithm · Regression | convergence rate: 1-Θ(1/K) -> 1-Θ(1/√K) | Helped |
| Use Heavy Ball momentum instead of vanilla GD | GNN learning algorithm · Binary classification | convergence rate: 1-Θ(1/K^2) -> 1-Θ(1/K) | Helped |
| Set momentum β to a small nonzero value rather than β=0 (ablation) | AGD for regression · Regression | convergence rate: ν(0)≥ 1-(1-ε₀)/(88κ²γK); ν(β*)= 1-(1-ε₀)/√(88κ²γK) | Helped |
| Set momentum β to a small nonzero value rather than β=0 (ablation) | AGD for binary classification · Binary classification | convergence rate: ν(0)= 1-(1-ε₀)/(11κ²γK²); ν(β*)= 1-√((1-ε₀)/(11κ²γK²)) | Helped |
| Use tensor initialization rather than the tested random initialization (ablation) | GNN learning algorithm · Synthetic graph-structured data | estimation performance | No clear effect |
| Increase maximum graph degree | GNN learning algorithm · Synthetic regression data | samples required for successful estimation | Hurt |
| Increase average node degree while holding maximum degree fixed | GNN learning algorithm · Random bounded-degree graph | samples required for successful recovery | Hurt |
| Change graph structure while keeping maximum degree and normalized-adjacency singular value similar | GNN learning algorithm · Connected cycle, two-dimensional grid, and random regular graph | samples required for accurate estimation | No clear effect |
Closest papers by meaning
- On the Interplay between Graph Structure and Learning Algorithms in Graph Neural Networks (ICML 2025 · similarity 0.91)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.90)
- A Manifold Perspective on the Statistical Generalization of Graph Neural Networks (ICML 2025 · similarity 0.89)
- Learning to Reweight for Generalizable Graph Neural Network (AAAI 2024 · similarity 0.89)
- Making Classic GNNs Strong Baselines Across Varying Homophily: A Smoothness–Generalization Perspective (NeurIPS 2025 · similarity 0.89)
- Generalization of Graph Neural Networks Is Robust to Model Mismatch (AAAI 2025 · 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.