Research map / Label-efficient and robust learning / Recognition and long-tail
Hub-Pathway: Transfer Learning from A Hub of Pre-trained Models
Yang Shu, Zhangjie Cao, Ziyang Zhang, Jianmin Wang, Mingsheng Long
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 |
|---|---|---|---|
| replace learned pathway with random pathway (ablation) | Hub-Pathway · COCO-70 | accuracy: 84.36 -> 80.97 | Hurt |
| replace learned pathway with random pathway (ablation) | Hub-Pathway · Stanford Cars | accuracy: 91.72 -> 90.61 | Hurt |
| remove explore loss (ablation) | Hub-Pathway · COCO-70 | accuracy: 84.36 -> 81.34 | Hurt |
| remove explore loss (ablation) | Hub-Pathway · Stanford Cars | accuracy: 91.72 -> 89.75 | Hurt |
| remove exploit loss (ablation) | Hub-Pathway · COCO-70 | accuracy: 84.36 -> 80.10 | Hurt |
| remove exploit loss (ablation) | Hub-Pathway · Stanford Cars | accuracy: 91.72 -> 90.59 | Hurt |
| data-dependent pathway instead of best single model (ablation) | Hub-Pathway · Average (homogeneous hub) | accuracy: 83.58 -> 85.63 | Improved |
| data-dependent pathway instead of task-level top-2 ensemble (ablation) | Hub-Pathway · Average (homogeneous hub) | accuracy: 84.31 -> 85.63 | Improved |
| per-datum oracle model selection vs learned pathway (ablation) | Hub-Pathway · Average (homogeneous hub) | accuracy: 85.63 -> 90.48 | Improved |
| Hub-Pathway instead of full Ensemble (ablation) | Hub-Pathway · CIFAR-100 (homogeneous hub) | accuracy: 84.50 -> 85.63 | Improved |
| Hub-Pathway instead of ensemble: FLOPs (ablation) | Hub-Pathway · CIFAR-100 | FLOPs (G): 20.55 -> 9.11 | Improved |
| Hub-Pathway instead of ensemble: inference speed (ablation) | Hub-Pathway · CIFAR-100 | speed (samples/s): 98.64 -> 240.48 | Improved |
Closest papers by meaning
- Transferring Knowledge across Learning Processes (ICLR 2019 · similarity 0.85)
- Fantastic Gains and Where to Find Them: On the Existence and Prospect of General Knowledge Transfer between Any Pretrained Model (ICLR 2024 · similarity 0.85)
- Knowledge Transfer from Vision Foundation Models for Efficient Training of Small Task-specific Models (ICML 2024 · similarity 0.85)
- MergeNet: Knowledge Migration Across Heterogeneous Models, Tasks, and Modalities (AAAI 2025 · similarity 0.84)
- Transferring Learning Trajectories of Neural Networks (ICLR 2024 · similarity 0.82)
- Diffusion Guided Chain-of-Vision for Large Autoregressive Vision Models (CVPR 2026 · similarity 0.82)
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.
