Research map / Graph learning / Graph self-supervised learning
Learning Pseudo-Riemannian Manifolds for Heterophilic Graphs via Graph Signature
Yun Young Choi, Asung Kil, Sun Woo Park, Minho Lee, Seokhwan Kim
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes learning pseudo-Riemannian manifolds (indefinite metric signatures) for graph neural networks to better model heterophilic graphs, where edges signify dissimilarity.
Problem
GNNs implicitly assume a Riemannian (positive-definite metric) data manifold, creating a geometric mismatch for heterophilic graphs where edges often encode dissimilarity or structural repulsion.
Approach
Shift from Riemannian to pseudo-Riemannian geometry, learning the metric signature of the manifold via a graph signature so that the geometry can represent repulsive (dissimilar) relations alongside attractive ones.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| allow indefinite metric (vs positive-definite Lambda) (ablation) | PRAT · 8 standard benchmarks, average | accuracy: 68.29 -> 71.96 | Improved |
| allow indefinite metric (vs positive-definite Lambda) (ablation) | PRAT · Cornell | accuracy: 57.58 -> 71.89 | Improved |
| allow indefinite metric (vs positive-definite Lambda) (ablation) | PRAT · Texas | accuracy: 76.78 -> 84.32 | Improved |
| allow indefinite metric (vs positive-definite Lambda) (ablation) | PRAT · Chameleon | accuracy: 65.64 -> 68.73 | Improved |
| allow indefinite metric (vs positive-definite Lambda) (ablation) | PRAT · Cora, PubMed, Citeseer | accuracy: Cora 88.42 vs 88.18; Citeseer 78.06 vs 78.09 | No clear effect |
| learn Lambda instead of freezing at signature initialization (ablation) | PRAT · Cornell | accuracy: 55.15 -> 71.89 | Improved |
| learn Lambda instead of freezing at signature initialization (ablation) | PRAT · Texas | accuracy: 74.89 -> 84.32 | Improved |
| learn Lambda instead of freezing it (ablation) | PRAT · large Platonov graphs | accuracy | No clear effect |
| indefinite metric vs signed aggregation weights only (FAGCN) on synthetic graphs | PRAT · synthetic graphs, h_E <= 0.4 | accuracy: gap averages 40.6 points | Improved |
| PRAT vs GAT and GATv2 under varying graph signature at h_E=0.5 | PRAT · synthetic graphs | accuracy std across signatures: GAT 3.99, GATv2 3.42, PRAT 0.68 | Improved |
| delete negative directions of signature but keep magnitudes (ablation) | PRAT · roman-empire | accuracy: -2.26 points | Hurt |
| replace signature with sign-only, sign-flipped or random draws (ablation) | PRAT · roman-empire | accuracy: within 0.7 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- BernNet — BernNet: Learning Arbitrary Graph Spectral Filters via Bernstein Approximation (NeurIPS 2021) · better on 1 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Pseudo-Riemannian Graph Transformer (NeurIPS 2025 · similarity 0.91)
- Motif-Aware Riemannian Graph Neural Network with Generative-Contrastive Learning (AAAI 2024 · similarity 0.89)
- Toward a Unified Geometry Understanding : Riemannian Diffusion Framework for Graph Generation and Prediction (NeurIPS 2025 · similarity 0.89)
- Spectro-Riemannian Graph Neural Networks (ICLR 2025 · similarity 0.88)
- Deeper with Riemannian Geometry: Overcoming Oversmoothing and Oversquashing for Graph Foundation Models (NeurIPS 2025 · similarity 0.86)
- Geometry Awakening: Cross-Geometry Learning Exhibits Superiority over Individual Structures (NeurIPS 2024 · 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.
