Research map / Graph learning / GNN architectures
Large-Scale Graph Neural Architecture Search
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 |
|---|---|---|---|
| add architecture peer learning on graphs (APLG) (ablation) | GAUSS supernet training · CS and Physics | Kendall tau [%]: 23.71 -> 64.86 (CS); 20.35 -> 53.51 (Physics) | Helped |
| add architecture importance sampling (AIS) on top of APLG (ablation) | GAUSS supernet training · CS and Physics | Kendall tau [%]: 64.86 -> 72.83 (CS); 53.51 -> 60.72 (Physics) | Helped |
| interaction within learning team with alpha_min below 1 (e.g., 0.6) (ablation) | GAUSS APLG · CS and Physics | Kendall tau [%]: alpha_min=1.0 gives 63.27 (CS) / 57.28 (Physics); best 0.6 gives 72.83 / 60.72; alpha_min=0.0 gives 63.25 / 51.13 | Mixed |
| increase learning team size from 2 to 10 (ablation) | GAUSS APLG · CS and Physics | Kendall tau: rises up to team size 10, then stabilizes or drops beyond 10 | Helped |
| search full GAUSS architectures vs. hand-crafted and NAS baselines | GNN architecture search on Arxiv, Products, Papers100M · Arxiv / Products / Papers100M | test accuracy [%]: Arxiv 72.35 vs DARTS 72.10; Products 81.26 vs SAGE 80.19; Papers100M 67.32 vs GCN 67.06 | Helped |
| GAUSS architecture combined with Node2Vec, label reuse, C&S, KD (ablation) | GAUSS vs GCN on Arxiv validation · Arxiv validation | validation accuracy [%]: GAUSS stays ahead of GCN at each stage (e.g., 74.55 vs 73.96 with all techniques) | Helped |
| GAUSS joint sampling vs. GNAS baselines on Products and Papers100M | GraphNAS, SGAS, DARTS, EGAN · Products / Papers100M | feasibility: baselines OOT or OOM on Products and Papers100M | No clear effect |
Closest papers by meaning
- Rethinking Graph Neural Architecture Search From Message-Passing (CVPR 2021 · similarity 0.91)
- Graph HyperNetworks for Neural Architecture Search (ICLR 2019 · similarity 0.90)
- Disentangled Continual Graph Neural Architecture Search with Invariant Modular Supernet (ICML 2024 · similarity 0.87)
- Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach (CVPR 2024 · similarity 0.87)
- MuseGNN: Forming Scalable, Convergent GNN Layers that Minimize a Sampling-Based Energy (ICLR 2025 · similarity 0.87)
- AutoGFM: Automated Graph Foundation Model with Adaptive Architecture Customization (ICML 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.