Research map / Label-efficient and robust learning / Recognition and long-tail
Not All Models Are Equal: Predicting Model Transferability in a Self-Challenging Fisher Space
Wenqi Shao, Xun Zhao, Yixiao Ge, Zhaoyang Zhang, Lei Yang, Xiaogang Wang, Ying Shan, Ping Luo
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 Reg-FDA instead of LogME as the scorer under ConfMix (ablation) | SFDA · Aircraft (supervised CNN pool) | weighted Kendall tau_w: 0.408 -> 0.615 | Improved |
| use Reg-FDA instead of LogME as the scorer under ConfMix (ablation) | SFDA · Flowers (supervised CNN pool) | weighted Kendall tau_w: 0.023 -> 0.542 | Improved |
| add ConfMix self-challenging (ablation) | SFDA · Aircraft (supervised CNN pool) | weighted Kendall tau_w: 0.481 -> 0.615 | Improved |
| add ConfMix self-challenging (ablation) | SFDA · Flowers (supervised CNN pool) | weighted Kendall tau_w: 0.668 -> 0.542 | Hurt |
| add ConfMix self-challenging (ablation) | SFDA · CF-10 (supervised CNN pool) | weighted Kendall tau_w: 0.887 -> 0.949 | Improved |
| use adaptive lambda instead of fixed lambda=0.5 (ablation) | SFDA · Aircraft (supervised CNN pool) | weighted Kendall tau_w: 0.424 -> 0.615 | Improved |
| use adaptive lambda instead of fixed lambda=0.5 (ablation) | SFDA · Flowers (supervised CNN pool) | weighted Kendall tau_w: 0.228 -> 0.542 | Improved |
| SFDA instead of NLEEP | SFDA · Average over 11 target tasks (self-supervised CNN pool) | weighted Kendall tau_w: 0.397 -> 0.593 | Improved |
| SFDA instead of NLEEP | SFDA · Average over 11 target tasks (ViT pool) | weighted Kendall tau_w: 0.218 -> 0.647 | Improved |
| SFDA instead of LogME | SFDA · Average over 11 target tasks (ViT pool) | weighted Kendall tau_w: 0.556 -> 0.647 | Improved |
| SFDA instead of NLEEP | SFDA · DTD (supervised CNN pool) | weighted Kendall tau_w: 0.777 -> 0.597 | Hurt |
| SFDA instead of re-training the head | SFDA · Aircraft (self-supervised CNN pool) | weighted Kendall tau_w: -0.008 -> 0.254 | 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.
- LogME — LogME: Practical Assessment of Pre-trained Models for Transfer Learning (ICML 2021) · 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
- Fast and Accurate Transferability Measurement by Evaluating Intra-class Feature Variance (ICCV 2023 · similarity 0.87)
- PACTran: PAC-Bayesian Metrics for Estimating the Transferability of Pretrained Models to Classification Tasks (ECCV 2022 · similarity 0.86)
- Building a Winning Team: Selecting Source Model Ensembles using a Submodular Transferability Estimation Approach (ICCV 2023 · similarity 0.85)
- Implicit Modeling for Transferability Estimation of Vision Foundation Models (NeurIPS 2025 · similarity 0.85)
- Features are fate: a theory of transfer learning in high-dimensional regression (ICML 2025 · 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.
