Research map / Graph learning / GNN architectures
LD2: Scalable Heterophilous Graph Neural Network with Decoupled Embeddings
Ningyi Liao, Siqiang Luo, Xiang Li, Jieming Shi
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 LD2 decoupled precomputation instead of LINKX | LD2 · genius | test accuracy (%): 82.51 -> 85.31 | Improved |
| use LD2 instead of LINKX | LD2 · pokec | test accuracy (%): 68.64 -> 74.93 | Improved |
| use LD2 instead of LINKX | LD2 · snap-patents | test accuracy (%): 52.69 -> 58.58 | Improved |
| use LD2 instead of LINKX | LD2 · penn94 | test accuracy (%): 78.63 -> 75.52 | Hurt |
| use LD2 instead of LINKX | LD2 · arxiv-year | test accuracy (%): 50.44 -> 50.29 | No clear effect |
| add graph topology embeddings on top of node features (LD2 vs MLP) | LD2 · snap-patents | test accuracy (%): 23.03 -> 58.58 | Improved |
| use LD2 instead of LINKX for GPU memory | LD2 · snap-patents | peak GPU memory (GB): 21.53 -> 3.96 | Improved |
| use LD2 instead of LINKX for GPU memory | LD2 · wiki | peak GPU memory (GB): 14.53 -> 4.47 | Improved |
| decoupled precomputation vs heterophilous baselines in training time | LD2 · large graphs | training time: 3-15 times faster | Improved |
| apply GSAINT minibatch sampling to MixHop | MixHop-GS · snap-patents | test accuracy (%): 34.73 vs LD2 58.58 | Hurt |
| add feature embeddings to raw attributes on genius (ablation) | LD2 channels · genius | test accuracy (%) | Improved |
| use only adjacency embedding PA vs integrating with other channels (ablation) | LD2 channels · genius | test accuracy (%) | Improved |
Closest papers by meaning
- Understanding Heterophily for Graph Neural Networks (ICML 2024 · similarity 0.90)
- Generative Graph Pattern Machine (NeurIPS 2025 · similarity 0.90)
- HeterSEED: Semantics–Structure Decoupling for Heterogeneous Graph Learning under Heterophily (NeurIPS 2026 · similarity 0.89)
- When Hypergraph Meets Heterophily: New Benchmark Datasets and Baseline (AAAI 2025 · similarity 0.89)
- On the Impact of Feature Heterophily on Link Prediction with Graph Neural Networks (NeurIPS 2024 · similarity 0.89)
- Integrating Co-Training with Edge Discrimination to Enhance Graph Neural Networks Under Heterophily (AAAI 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.
