Research map / Graph learning / GNN architectures
LRIM: a Physics-Based Benchmark for Provably Evaluating Long-Range Capabilities in Graph Learning
Joël Mathys, Henrik Christiansen, Federico Errica, Takashi Maruyama, Francesco Alesiani
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 |
|---|---|---|---|
| lower sigma from 1.5 to 0.6 (harder variant) (ablation) | oracle predictor · LRIM oracle r-hop restriction | LogMSE | Hurt |
| increase system size L (ablation) | oracle predictor · LRIM oracle r-hop restriction | LogMSE | Hurt |
| add more layers up to about 10 and 16 (ablation) | MPNN and GPS baselines · LRIM-16-hard and LRIM-32-hard | LogMSE | Improved |
| add layers beyond roughly 10 (LRIM-16) and 16 (LRIM-32) (ablation) | MPNN and GPS baselines · LRIM-16-hard and LRIM-32-hard | LogMSE | No clear effect |
| use graph transformer GPS-Base instead of GatedGCN | baseline · LRIM-16-hard | LogMSE: -3.844 -> -4.211 | Improved |
| use GPS-Base instead of GatedGCN | baseline · LRIM-32-hard | LogMSE: -4.087 -> -4.044 | No clear effect |
| add virtual node to GatedGCN | GatedGCN · LRIM-16-hard | LogMSE: -3.844 -> -4.068 | Improved |
| add virtual node to GatedGCN | GatedGCN · LRIM-32-hard | LogMSE: -4.087 -> -3.243 | Hurt |
| add RWSE encoding to GPS | GPS-Base · LRIM-16-hard | LogMSE: -4.211 -> -4.011 | Hurt |
| add LapPE encoding to GPS | GPS-Base · LRIM-16-hard | LogMSE: -4.211 -> -4.334 | Improved |
| transfer LRIM-16-hard model to LRIM-32-hard without retraining | GPS-Base · LRIM-32-hard | LogMSE: -4.340 -> -1.057 | Hurt |
| transfer LRIM-16-hard model to LRIM-32-hard without retraining | GatedGCN · LRIM-32-hard | LogMSE: -3.919 -> -1.050 | Hurt |
Closest papers by meaning
- On Measuring Long-Range Interactions in Graph Neural Networks (ICML 2025 · similarity 0.92)
- Get a GRIP, this will be a long TRIP: A Quantifiable Long-Range Framework for Verifying Over-squashing (NeurIPS 2026 · similarity 0.90)
- Learning Long Range Dependencies on Graphs via Random Walks (ICLR 2025 · similarity 0.88)
- No Metric to Rule Them All: Toward Principled Evaluations of Graph-Learning Datasets (ICML 2025 · similarity 0.88)
- How Do Large Language Models Understand Graph Patterns? A Benchmark for Graph Pattern Comprehension (ICLR 2025 · similarity 0.87)
- Adaptive Message Passing: A General Framework to Mitigate Oversmoothing, Oversquashing, and Underreaching (ICML 2025 · similarity 0.87)
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.
