Research map / Label-efficient and robust learning / Recognition and long-tail
Beyond Model Base Retrieval: Weaving Knowledge to Master Fine-grained Neural Network Design
Jialiang Wang, Hanmo Liu, Shimin Di, Zhili Wang, Jiachuan Wang, Lei Chen, Xiaofang Zhou
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 |
|---|---|---|---|
| add dynamic (Bayesian) task similarity update (ablation) | M-DESIGN · Node classification, Kendall correlation | avg Kendall correlation: 0.08 -> 0.34 | Improved |
| add dynamic task similarity update (ablation) | M-DESIGN · Node classification, Actor | accuracy: 33.71 -> 34.89 | Improved |
| add sliding windows to control locality (ablation) | M-DESIGN · Node classification, Kendall correlation | avg Kendall correlation: 0.27 -> 0.34 | Improved |
| add sliding windows (ablation) | M-DESIGN · Node classification, Cornell | accuracy: 75.68 -> 77.48 | Improved |
| add OOD adaptation with predictive task planners (ablation) | M-DESIGN · Node classification, Computers/Cornell/Texas | accuracy: Cornell 75.68 -> 77.48; Texas 81.98 -> 83.79 | Improved |
| add OOD adaptation (ablation) | M-DESIGN · Node classification, Photo/CS/PubMed | accuracy: Photo 94.75 vs 94.62; CS 95.33 vs 95.16 | No clear effect |
| enable OOD adaptation in gain transfer from closest benchmark (ablation) | M-DESIGN · Cornell, gain correlation R2 | R-squared: 0.03 -> 0.11 | Improved |
| reduce MKB to 25% of benchmark sources (ablation) | M-DESIGN · Node classification (avg of 11) | accuracy: 82.11 -> 81.50 | Hurt |
| use different similarity initialization (weighted avg, majority vote, most similar) (ablation) | M-DESIGN · Node classification (avg of 11) | accuracy: 81.80 / 81.79 / 81.31 | No clear effect |
| use M-DESIGN refinement vs best baseline DesiGNN | Node classification · Node classification, Texas | accuracy: 81.80 -> 83.79 | Improved |
| add refinement on top of one-shot retrieval in image domain | M-DESIGN-Init · NATS-Bench | accuracy: Cifar10 0.8896 vs optimal 0.8916 | No clear effect |
| use M-DESIGN simplified variant for tabular tasks | Our-S · HPOBench Naval | MSE: 0.000453 -> 0.000034 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AutoTransfer — Auto-Transfer: Learning to Route Transferable Representations (ICLR 2022) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Design Principle Transfer in Neural Architecture Search via Large Language Models (AAAI 2025 · similarity 0.85)
- Structuring Open-Ended NAS: Semi-Automated Design Knowledge Structuring with LLMs for Efficient Neural Architecture Search (NeurIPS 2026 · similarity 0.85)
- WISE: Rethinking the Knowledge Memory for Lifelong Model Editing of Large Language Models (NeurIPS 2024 · similarity 0.84)
- An Architecture Search Framework for Inference-Time Techniques (ICML 2025 · similarity 0.84)
- Know2Vec: A Black-Box Proxy for Neural Network Retrieval (AAAI 2025 · similarity 0.84)
- Neural Architecture Retrieval (ICLR 2024 · similarity 0.84)
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.
