Research map / Graph learning / GNN architectures
Descent Steps of a Relation-Aware Energy Produce Heterogeneous Graph Neural Networks
Hongjoon Ahn, Yongyi Yang, Quan Gan, Taesup Moon, David Wipf
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 |
|---|---|---|---|
| Replace fixed identity compatibility matrix with trainable H_t (ablation) | HALO · MUTAG | accuracy (%): 77.94 -> 86.17 | Improved |
| Replace fixed identity compatibility matrix with trainable H_t (ablation) | HALO · BGS | accuracy (%): 86.90 -> 93.10 | Improved |
| Replace fixed identity compatibility matrix with trainable H_t (ablation) | HALO · DBLP | accuracy (%): 95.40 -> 96.30 | Improved |
| Add proximal gradient descent (non-linear activation) instead of plain gradient descent (ablation) | HALO · MUTAG | accuracy (%): 80.88 -> 86.17 | Improved |
| Add proximal gradient descent instead of plain gradient descent (ablation) | HALO · BGS | accuracy (%): 86.17 -> 93.10 | Improved |
| Add proximal gradient descent instead of plain gradient descent (ablation) | HALO · DBLP | accuracy (%): 95.90 -> 96.30 | Improved |
| Use 2-layer MLP instead of linear base function f (ablation) | HALO · MUTAG, BGS, DBLP | accuracy (%): 86.17 -> 85.58 (MUTAG), 93.10 -> 93.79 (BGS), 96.30 -> 94.75 (DBLP) | Mixed |
| Use trainable compatibility matrices and node features instead of fixed matrices and labels only (ZooBP) | HALO · DBLP-reduced k=20 | accuracy (%) all nodes: 28.47 -> 43.00 | Improved |
| Use HALO instead of ZooBP | HALO · Academic-reduced k=18 | accuracy (%) all nodes: 42.94 -> 90.18 | Improved |
| Use HALO instead of ZooBP | HALO · Academic-reduced k=4 | accuracy (%) all nodes: 75.97 -> 94.93 | Improved |
| Use HALO instead of Simple-HGN | HALO · Freebase | accuracy (%): 66.29 -> 62.06 | Hurt |
| Use HALO instead of Simple-HGN | HALO · IMDB | accuracy (%): 67.36 -> 76.20 | Improved |
Closest papers by meaning
- Heterophily-Aware Representation Learning on Heterogeneous Graphs (TPAMI 2025 · similarity 0.92)
- HeterSEED: Semantics–Structure Decoupling for Heterogeneous Graph Learning under Heterophily (NeurIPS 2026 · similarity 0.91)
- Understanding Heterophily for Graph Neural Networks (ICML 2024 · similarity 0.89)
- On the Impact of Feature Heterophily on Link Prediction with Graph Neural Networks (NeurIPS 2024 · similarity 0.89)
- Heterogeneous Graph Neural Network on Semantic Tree (AAAI 2025 · 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.
