Research map / Label-efficient and robust learning / Domain adaptation and generalization
Holistic Transfer: Towards Non-Disruptive Fine-Tuning with Partial Target Data
Cheng-Hao Tu, Hong-You Chen, Zheda Mai, Jike Zhong, Vardaan Pahuja, Tanya Berger-Wolf, Song Gao, Charles Stewart, Yu Su, Wei-Lun Chao
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 |
|---|---|---|---|
| freeze the linear classifier during fine-tuning (ablation) | Naive Target / SGD · Office-Home Avg | unseen accuracy (%): 22.36 -> 43.44 | Improved |
| add selective distillation loss on unseen logits (ablation) | SGD (frozen) · Office-Home Avg | unseen accuracy (%): 43.44 -> 57.41 | Improved |
| add feature rank regularization (ablation) | SGD (frozen) · Office-Home Avg | unseen accuracy (%): 43.44 -> 55.64 | Improved |
| replace SGD with leave-out local SGD (LOLSGD) (ablation) | SGD (frozen) · Office-Home Avg | unseen accuracy (%): 43.44 -> 53.98 | Improved |
| replace SWAD with LOLSGD (ablation) | SWAD (frozen) · Office-Home Avg | unseen accuracy (%): 45.30 -> 53.98 | Improved |
| add distillation and rank regularizers to LOLSGD (ablation) | LOLSGD (frozen) · Office-Home Avg | unseen accuracy (%): 53.98 -> 66.54 | Improved |
| update only batchnorm statistics instead of naive fine-tuning | Naive Target · Office-Home Avg | unseen accuracy (%): 22.36 -> 67.51 | Improved |
| apply LP-FT instead of naive fine-tuning | Naive Target · Office-Home Avg | unseen accuracy (%): 22.36 -> 15.60 | Hurt |
| ensemble predictions with source model (SE) (ablation) | Naive Target · Office-Home Avg | overall accuracy (%): 53.50 -> 67.46 | Improved |
| add source ensemble to LOLSGD + distill (ablation) | LOLSGD+distill · FEMNIST | overall accuracy (%): 85.76 -> 85.10 | Hurt |
| replace SGD with LOLSGD (ablation) | SGD (frozen) · FEMNIST | unseen accuracy (%): 64.00 -> 66.27 | Mixed |
| replace SGD with LOLSGD (ablation) | SGD (frozen) · iWildCam | unseen accuracy (%): 7.91 -> 13.30 | Improved |
Closest papers by meaning
- Rethinking the Role of Pre-Trained Networks in Source-Free Domain Adaptation (ICCV 2023 · similarity 0.88)
- Co-Tuning for Transfer Learning (NeurIPS 2020 · similarity 0.87)
- Active Universal Domain Adaptation (ICCV 2021 · similarity 0.87)
- Universal Semi-Supervised Domain Adaptation by Mitigating Common-Class Bias (CVPR 2024 · similarity 0.87)
- Enhancing Semi-supervised Domain Adaptation via Effective Target Labeling (AAAI 2024 · similarity 0.86)
- Diffusion-Driven Progressive Target Manipulation for Source-Free Domain Adaptation (NeurIPS 2025 · similarity 0.86)
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.
