Research map / Graph learning / GNN architectures
Expressive and Invariant Graph Learning via Canonical Tree Cover Neural Networks
Michael Ito, Danai Koutra, Jenna Wiens
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 a spanning tree cover instead of a single tree (ablation) | CTNN · SCOP | accuracy: 68.7 -> 72.1 | Improved |
| use a spanning tree cover instead of a single tree (ablation) | CTNN · GO BIO | AUC: 69.4 -> 82.0 | Improved |
| use a spanning tree cover instead of a single tree (ablation) | CTNN · GO MOL | AUC: 79.9 -> 86.6 | Improved |
| use a spanning tree cover instead of a single tree (ablation) | CTNN · BBBP | AUC: 87.4 -> 86.1 | No clear effect |
| use TreeMPNN instead of plain MPNN on trees (ablation) | CTNN · SCOP | accuracy: 57.1 -> 72.1 | Improved |
| use TreeMPNN instead of plain MPNN on trees (ablation) | CTNN · HIV | AUC: 73.8 -> 75.2 | Improved |
| use TreeMPNN instead of plain MPNN on trees (ablation) | CTNN · BBBP | AUC: 81.3 -> 86.1 | Improved |
| use TreeMPNN (residual-edge message passing) instead of TreeRNN (ablation) | CTNN · SCOP | accuracy: 64.8 -> 72.1 | Improved |
| use TreeMPNN instead of TreeRNN (ablation) | CTNN · BACE | AUC: 83.5 -> 79.3 | Hurt |
| use TreeMPNN instead of TreeRNN (ablation) | CTNN · HIV | AUC: 77.9 -> 75.2 | Hurt |
| use CTNN tree cover instead of RCM sequence ordering | CTNN · ClinTox | AUC: 70.7 -> 84.7 | Improved |
| use CTNN instead of hand-crafted Fingerprint | CTNN · BACE | AUC: 82.9 -> 79.3 | Hurt |
Closest papers by meaning
- Neural Trees for Learning on Graphs (NeurIPS 2021 · similarity 0.88)
- Rethinking the Power of Graph Canonization in Graph Representation Learning with Stability (ICLR 2024 · similarity 0.88)
- Homomorphism Expressivity of Spectral Invariant Graph Neural Networks (ICLR 2025 · similarity 0.87)
- On the Expressive Power of Spectral Invariant Graph Neural Networks (ICML 2024 · similarity 0.86)
- Exploring Consistency in Graph Representations: from Graph Kernels to Graph Neural Networks (NeurIPS 2024 · similarity 0.86)
- On the Rademacher Complexity of Graph Neural Networks: Unifying Expressivity and Geometry (NeurIPS 2026 · similarity 0.86)
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.
