Research map / Graph learning / GNN architectures
Neural Execution of Graph Algorithms
Petar Veličković, Rex Ying, Matilde Padovano, Raia Hadsell, Charles Blundell
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 max aggregation rather than mean or sum aggregation | MPNN graph algorithm executor · Shortest-path predecessor prediction on graphs of 20–100 nodes | last-step accuracy: 93.20% -> 96.84%; 58.02% -> 93.88%; 43.93% -> 88.79% | Improved |
| Use max aggregation rather than mean aggregation | MPNN graph algorithm executor · Prim next-node selection, 20 nodes | accuracy: 90.56% -> 87.85% | Hurt |
| Use max aggregation rather than mean aggregation | MPNN graph algorithm executor · Prim next-node selection, 50 and 100 nodes | accuracy: 52.23% -> 63.89%; 20.63% -> 41.37% | Improved |
| Add simultaneous reachability training (ablation) | Bellman-Ford shortest-path predecessor learning · Shortest-path predecessor, 100 nodes | last-step accuracy: 81.06% -> 88.79% | Improved |
| Add intermediate algorithm supervision, including distance information (ablation) | MPNN-max shortest-path learning · Shortest-path predecessor, 100 nodes | last-step accuracy: 78.84% -> 88.79% | Improved |
| Train reachability first, then fine-tune on Bellman-Ford instead of learning both simultaneously (ablation) | MPNN-max shortest-path learning · Shortest-path predecessor, 100 nodes | last-step accuracy: 88.79% -> 83.16% | Hurt |
| Add sequential node selection to Prim execution rather than predicting predecessors directly (ablation) | MPNN-max Prim executor · Prim MST predecessor, 100 nodes | accuracy: 23.61% -> 90.02% | Improved |
Closest papers by meaning
- What do Graph Neural Networks learn? Insights from Tropical Geometry (NeurIPS 2024 · similarity 0.90)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.90)
- On the Interplay between Graph Structure and Learning Algorithms in Graph Neural Networks (ICML 2025 · similarity 0.89)
- Are Graph Neural Networks Optimal Approximation Algorithms? (NeurIPS 2024 · similarity 0.89)
- On the Power of Small-size Graph Neural Networks for Linear Programming (NeurIPS 2024 · similarity 0.89)
- Expressive Power of Graph Neural Networks for (Mixed-Integer) Quadratic Programs (ICML 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.
