Research map / Graph learning / Graph algorithms and clustering
AutoTransfer: AutoML with Knowledge Transfer - An Application to 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 |
|---|---|---|---|
| transfer task-informed design priors rather than use uniform random sampling | AutoTransfer · CoauthorPhysics, CoraFull, OGB-Arxiv, COX2, IMDB, PROTEINS | test accuracy: 95.16±0.55 -> 96.64±0.42 (CoauthorPhysics); 61.24±4.04 -> 69.27±0.76 (CoraFull); 67.92±1.92 -> 71.42±0.39 (OGB-Arxiv); 7 | Helped |
| use transferred design priors with 30 trials rather than TPE with 30 trials | AutoTransfer · CoauthorPhysics, CoraFull, OGB-Arxiv, COX2, IMDB, PROTEINS | test accuracy: 96.41±0.36 -> 96.91±0.27; 66.37±1.73 -> 70.05±0.42; 71.35±0.44 -> 72.21±0.27; 82.27±2.01 -> 86.52±1.58; 50.33±4.00 -> 54 | Mixed |
| use normalized losses instead of FIM-based task features and learned embeddings (ablation) | AutoTransfer · OGB-Arxiv | test accuracy: 71.42±0.39 -> 68.13±1.27 | Hurt |
| add the learned projection g to task features (ablation) | AutoTransfer · OGB-Arxiv | test accuracy: 70.67±0.52 -> 71.42±0.39 | Helped |
| add the learned projection g to task features (ablation) | AutoTransfer · task similarity rankings | Kendall rank correlation: 0.18±0.30 -> 0.43±0.22 | Helped |
| increase anchor models from 3 to 12 for task features (ablation) | AutoTransfer · OGB-Arxiv (3 trials) | test accuracy: 69.42±0.82 -> 70.67±0.52 | Helped |
| increase anchor models from 3 to 12 for task embeddings (ablation) | AutoTransfer · OGB-Arxiv (3 trials) | test accuracy: 69.80± 0.75 -> 71.42±0.39 | Helped |
| increase anchor models from 3 to 12 for task embeddings (ablation) | AutoTransfer · task similarity rankings | Kendall rank correlation: 0.12±0.28 -> 0.43±0.22 | Helped |
| use learned task embeddings instead of unprojected task features (ablation) | AutoTransfer · OGB-Arxiv (3 trials) | test accuracy: 70.67±0.52 -> 71.42±0.39 | Helped |
| use learned task embeddings instead of unprojected task features (ablation) | AutoTransfer · task similarity rankings | Kendall rank correlation: 0.18±0.30 -> 0.43±0.22 | Helped |
| use 3 anchor models rather than 12 (ablation) | AutoTransfer task embedding · task similarity rankings | Kendall rank correlation: 0.43±0.22 -> 0.12±0.28 | Hurt |
| use 3 anchor models rather than 12 (ablation) | AutoTransfer task embedding · OGB-Arxiv (3 trials) | test accuracy: 71.42±0.39 -> 69.80± 0.75 | Hurt |
Closest papers by meaning
- AutoGFM: Automated Graph Foundation Model with Adaptive Architecture Customization (ICML 2025 · similarity 0.87)
- AutoSGNN: Automatic Propagation Mechanism Discovery for Spectral Graph Neural Networks (AAAI 2025 · similarity 0.86)
- Towards Graph Foundation Models: Learning Generalities Across Graphs via Task-Trees (ICML 2025 · similarity 0.85)
- Towards Pre-trained Graph Condensation via Optimal Transport (NeurIPS 2025 · similarity 0.85)
- HeterGP: Bridging Heterogeneity in Graph Neural Networks with Multi-View Prompting (AAAI 2025 · similarity 0.85)
- GraphBridge: Towards Arbitrary Transfer Learning in GNNs (ICLR 2025 · similarity 0.85)
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.