Research map / Label-efficient and robust learning / Domain adaptation and generalization
Learning to Transfer Learn: Reinforcement Learning-Based Selection for Adaptive Transfer Learning
Linchao Zhu, Sercan . Arık, Yi Yang, Tomas Pfister
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 L2TL RL-based source class weighting | fine-tuning · Birdsnap (ImageNet source) | top-1 accuracy: 77.1 -> 78.1 | Improved |
| add L2TL RL-based source class weighting | fine-tuning · Stanford Cars (ImageNet source) | top-1 accuracy: 92.0 -> 92.6 | Improved |
| add L2TL RL-based source class weighting | fine-tuning · FGVC Aircraft (ImageNet source) | top-1 accuracy: 88.2 -> 89.1 | Improved |
| use L2TL instead of DATL class-prior matching | DATL · Birdsnap (JFT-300M source) | top-1 accuracy: 81.7 -> 82.4 | Improved |
| add L2TL with large JFT-300M source | fine-tuning · Birdsnap (JFT-300M source) | top-1 accuracy: 74.9 -> 82.4 | Improved |
| add L2TL for MNIST to SVHN transfer | fine-tuning · SVHN (MNIST source, 600 train images) | top-1 accuracy: 71.7 -> 75.2 | Improved |
| add L2TL in low-shot regime (5 samples per class) | fine-tuning · Stanford Cars (5 samples per class) | top-1 accuracy: 40.0 -> 46.5 | Improved |
| add L2TL with 20 samples per class | fine-tuning · Stanford Cars (20 samples per class) | top-1 accuracy: 84.9 -> 86.1 | Improved |
| add L2TL on dissimilar-domain texture target | fine-tuning · DTD split 1 (full training set) | top-1 accuracy: 70.3 -> 72.0 | Improved |
| add L2TL on DTD with 10 examples per class | fine-tuning · DTD split 1 (10 examples per class) | top-1 accuracy: 55.0 -> 60.1 | Improved |
| add L2TL on chest X-ray target | fine-tuning · CheXpert (Mean) | AUC: 89.3 -> 90.0 | Improved |
| use ImageNet fine-tuning instead of random initialization | random initialization · DTD split 1 (full training set) | top-1 accuracy: 57.4 -> 70.3 | Improved |
Closest papers by meaning
- Model-Based Transfer Learning for Contextual Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- A High-Dimensional Statistical Method for Optimizing Transfer Quantities in Multi-Source Transfer Learning (NeurIPS 2025 · similarity 0.87)
- Reinforcement Learning Guided Semi-Supervised Learning (NeurIPS 2024 · similarity 0.86)
- Wasserstein Transfer Learning (NeurIPS 2025 · similarity 0.85)
- Mixed-Sample SGD: an End-to-end Analysis of Supervised Transfer Learning (NeurIPS 2025 · similarity 0.85)
- When is Transfer Learning Possible? (ICML 2024 · 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.
