Research map / Graph learning / GNN architectures
Graph Ordering Attention Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Learn an attention-based neighborhood ordering and aggregate the ordered sequence with an LSTM rather than use GraphSAGE's random-order LSTM | GraphSAGE-style neighborhood aggregation · Top-2 pooling | accuracy: 65.05 ± 8.71 -> 69.21 ± 5.10 | Helped |
| Use GOAT with an LSTM aggregator instead of the strongest listed baseline | Node structural-property prediction · Betweenness Centrality (N=1000, p=0.01) | MSE: 0.0020 $ \pm 0.0008$ -> 0.0006 $ \pm 0.0002$ | Helped |
| Use GOAT with an LSTM aggregator instead of the strongest listed baseline | Node classification · Cora | accuracy: 83.0 -> 84.9 | Helped |
| Fix the neighborhood ordering using the random epoch-0 ordering rather than learning it (ablation) | GOAT(lstm) · Cora | accuracy: 83.36 ± 0.42 -> 81.56 ± 0.19 | Hurt |
| Use the fixed ordering extracted at epoch 100 instead of jointly learning the ordering (ablation) | GOAT(lstm) · Cora | accuracy: 83.36 ± 0.42 -> 83.80 ± 1.14 | Helped |
| Replace GAT attention with GATv2 attention (ablation) | GOAT(lstm) · Cora | accuracy: 84.9 -> 83.1 | No clear effect |
| Vary the number of attention heads (ablation) | GOAT(lstm) · Cora | accuracy: 84.9 (1 head); 83.1 (2 heads); 82.8 (4 heads); 82.9 (8 heads) | Mixed |
| Vary the number of attention heads (ablation) | GOAT(lstm) · CiteSeer | accuracy: 67.9 (1 head); 68.2 (2 heads); 69.5 (4 heads); 68.8 (8 heads) | Mixed |
| Vary the number of attention heads (ablation) | GOAT(lstm) · Disease | accuracy: 89.14 ± 2.99 (1 head); 90.78 ± 1.93 (2 heads); 91.32 ± 2.71 (4 heads); 92.11 ± 1.88 (8 heads) | Mixed |
| Use a GRU rather than an LSTM as the sequence aggregator (ablation) | GOAT · CiteSeer | accuracy: 69.5 -> 70.0 | Helped |
| Use a standard RNN rather than an LSTM as the sequence aggregator (ablation) | GOAT · Cora | accuracy: 84.9 -> 84.2 | Hurt |
Closest papers by meaning
- Tailoring Self-Attention for Graph via Rooted Subtrees (NeurIPS 2023 · similarity 0.89)
- Understanding Attention and Generalization in Graph Neural Networks (NeurIPS 2019 · similarity 0.89)
- Towards Deep Attention in Graph Neural Networks: Problems and Remedies (ICML 2023 · similarity 0.89)
- HONGAT: Graph Attention Networks in the Presence of High-Order Neighbors (AAAI 2024 · similarity 0.88)
- GOAt: Explaining Graph Neural Networks via Graph Output Attribution (ICLR 2024 · similarity 0.88)
- Greener GRASS: Enhancing GNNs with Encoding, Rewiring, and Attention (ICLR 2025 · 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.