Research map / Graph learning / GNN architectures
Directed Acyclic Graph Neural 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 |
|---|---|---|---|
| Use DAGNN rather than GAT | OGBG-CODE TOK · OGBG-CODE TOK | F1: 33.59\std{0.32} -> 34.41\std{0.38} | Helped |
| Use DAGNN rather than GAT | OGBG-CODE-15 TOK · OGBG-CODE-15 TOK | F1: 27.37\std{0.16} -> 29.11\std{0.44} | Helped |
| Use DAGNN rather than D-VAE | OGBG-CODE LP · OGBG-CODE LP | accuracy: 99.90\std{0.02} -> 99.93\std{0.01} | Helped |
| Use DAGNN rather than D-VAE | NA latent representation · NA | RMSE: 0.375\std{0.003} -> 0.264\std{0.004} | Helped |
| Use DAGNN rather than D-VAE | BN latent representation · BN | RMSE: 0.281\std{0.004} -> 0.122\std{0.004} | Helped |
| Add attention aggregation rather than gated-sum aggregation (ablation) | DAGNN · TOK-15 | F1: 24.98\std{0.45} -> 29.11\std{0.44} | Helped |
| Add attention aggregation rather than gated-sum aggregation (ablation) | DAGNN · LP-15 | accuracy: 99.88\std{0.02} -> 99.86\std{0.04} | No clear effect |
| Add multiple layers rather than use one layer (ablation) | DAGNN · TOK-15 | F1: 28.39\std{0.80} -> 29.11\std{0.44} | Helped |
| Use GRU rather than a fully connected combine layer (ablation) | DAGNN · TOK-15 | F1: 26.08\std{0.80} -> 29.11\std{0.44} | Helped |
| Pool target nodes rather than all nodes (ablation) | DAGNN · BN | RMSE: 0.098\std{0.003} -> 0.122\std{0.004} | Hurt |
| Pool target nodes rather than all nodes (ablation) | DAGNN · NA | RMSE: 0.302\std{0.002} -> 0.264\std{0.004} | Helped |
| Retain edge attributes (ablation) | DAGNN · TOK-15 | F1: 28.85\std{0.24} -> 29.11\std{0.44} | Helped |
Closest papers by meaning
- Transformers over Directed Acyclic Graphs (NeurIPS 2023 · similarity 0.91)
- DAG-GNN: DAG Structure Learning with Graph Neural Networks (ICML 2019 · similarity 0.91)
- DAGs with NO TEARS: Continuous Optimization for Structure Learning (NeurIPS 2018 · similarity 0.89)
- Directed Graph Grammars for Sequence-based Learning (ICML 2025 · similarity 0.89)
- Commute Graph Neural Networks (ICML 2025 · similarity 0.89)
- NN-Former: Rethinking Graph Structure in Neural Architecture Representation (CVPR 2025 · similarity 0.88)
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.