Research map / Graph learning / GNN architectures
LEAP: Local ECT-Based Learnable Positional Encodings for Graphs
Juan P Garcia Amboage, Ernst Röell, Patrick Schnider, Bastian Rieck
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 |
|---|---|---|---|
| add LEAP positional encoding | GCN/GAT on synthetic edge-count task · Synthetic edge-count | accuracy: 71.83 -> 100.0 (GCN) | Improved |
| add LEAP-L PE to a GCN | GCN · Letter-H | accuracy: 41.6 -> 74.2 | Improved |
| add LEAP-L to NoMP versus RWPE | NoMP · Letter-H | accuracy: 79.2 -> 81.6 | Improved |
| use 2-hop instead of 1-hop neighborhoods (ablation) | LEAP-F · Letter-H | accuracy: 81.29 -> 74.44 | Hurt |
| concatenate 1-hop and 2-hop LEAP (ablation) | LEAP-F · Letter-H | accuracy: 81.29 -> 77.91 | Hurt |
| learn directions instead of fixing them (ablation) | LEAP (GCN, NoMP) · TU datasets | accuracy: e.g. GCN Letter-H 72.2 -> 74.2 | Improved |
| learn directions instead of fixing them (ablation) | LEAP with GAT · TU datasets | accuracy: e.g. COX2 79.2 -> 80.1 | Mixed |
| vary PE dimension 2/5/10/20 (ablation) | LEAP · Letter-H/M/L (NoMP) | accuracy: Letter-H LEAP-F 81.16 (dim 2) vs 80.84 (dim 20) | No clear effect |
| choice of projection strategy (linear, attention, DeepSets, 1D conv) (ablation) | LEAP · TU datasets | accuracy: GCN Letter-H LEAP-L 74.2 (linear) vs 62.7 (attn) | Mixed |
| combine LEAP with LaPE on HIV | GCN · HIV | AUROC | Improved |
| use LEAP instead of global DECT with CNN | graph classification · TU datasets | accuracy: Letter-H 85.4 (DECT 65K) vs 74.2 (LEAP-L GCN) | Mixed |
Closest papers by meaning
- Diss-l-ECT: Dissecting Graph Data with Local Euler Characteristic Transforms (ICML 2025 · similarity 0.93)
- Learning Efficient Positional Encodings with Graph Neural Networks (ICLR 2025 · similarity 0.91)
- Differentiable Euler Characteristic Transforms for Shape Classification (ICLR 2024 · similarity 0.89)
- On the Stability of Expressive Positional Encodings for Graphs (ICLR 2024 · similarity 0.89)
- Positional Encoding meets Persistent Homology on Graphs (ICML 2025 · similarity 0.89)
- Effective Structural Encodings via Local Curvature Profiles (ICLR 2024 · 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.
