Research map / Label-efficient and robust learning / Domain adaptation and generalization
Transfer Learning with Affine Model Transformation
Shunya Minami, Kenji Fukumizu, Yoshihiro Hayashi, Ryo Yoshida
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 |
|---|---|---|---|
| use affine model transfer on top of feature extraction with GPT-3 embeddings | feature-extraction TL · SciRepEval review score | RMSE: 1.3284 -> 1.3234 | Improved |
| use affine model transfer instead of feature extraction with SciBERT embeddings | feature-extraction TL · SciRepEval review score | RMSE: 1.2856 -> 1.2797 | Improved |
| use affine model transfer instead of feature extraction with BERT embeddings | feature-extraction TL · SciRepEval review score | RMSE: 1.3086 -> 1.3069 | Improved |
| use affine model transfer instead of feature extraction with GPT-3 embeddings | feature-extraction TL · SciRepEval acceptance | accuracy: 0.6279 -> 0.6386 | Improved |
| use affine model transfer instead of feature extraction with SciBERT embeddings | feature-extraction TL · SciRepEval acceptance | accuracy: 0.6520 -> 0.6507 | No clear effect |
| add source features via AffineTL-full vs Direct on a weakly related task | Direct kernel ridge · SARCOS Torque 1, n=5 | RMSE: 21.3 -> 21.2 | No clear effect |
| add source features via AffineTL-full vs Direct on a related task | Direct kernel ridge · SARCOS Torque 7, n=5 | RMSE: 2.66 -> 2.23 | Improved |
| use AffineTL-full instead of Direct at n=50 | Direct kernel ridge · SARCOS Torque 7, n=50 | RMSE: 1.05 -> 0.870 | Improved |
| use Only source features instead of Direct | Direct kernel ridge · SARCOS Torque 1, n=5 | RMSE: 21.3 -> 24.0 | Hurt |
| use HTL-offset instead of Direct (negative transfer) | Direct kernel ridge · SARCOS Torque 1, n=50 | RMSE: 10.8 -> 15.1 | Hurt |
| use Fine-tune instead of AffineTL-full when source-target relation is strong | AffineTL-full · SARCOS Torque 7, n=50 | RMSE: 0.870 -> 0.790 | Improved |
| use PAC-Net instead of Direct on a weakly related task, n=50 | Direct kernel ridge · SARCOS Torque 1, n=50 | RMSE: 10.8 -> 17.1 | Hurt |
Closest papers by meaning
- Transfer Learning of Linear Regression with Multiple Pretrained Models: Benefiting from More Pretrained Models via Overparameterization Debiasing (NeurIPS 2026 · similarity 0.86)
- Wasserstein Transfer Learning (NeurIPS 2025 · similarity 0.86)
- On Hypothesis Transfer Learning of Functional Linear Models (ICML 2024 · similarity 0.85)
- Transfer Learning for Benign Overfitting in High-Dimensional Linear Regression (NeurIPS 2025 · similarity 0.85)
- Transferring Knowledge From Large Foundation Models to Small Downstream Models (ICML 2024 · similarity 0.85)
- Universality in Transfer Learning for Linear Models (NeurIPS 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.
