Research map / Label-efficient and robust learning / Recognition and long-tail
LogME: Practical Assessment of Pre-trained Models for Transfer Learning
Kaichao You, Yong Liu, 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 |
|---|---|---|---|
| Use LogME instead of LEEP or NCE to rank supervised pre-trained models | Pre-trained model selection · Vision classification benchmarks | weighted Kendall tau: Aircraft: LogME 0.54 vs NCE 0.40; Caltech: 0.70 vs 0.65; Cars: 0.65 vs LEEP 0.41; CIFAR10: 0.78 vs LEEP 0.64; CIFAR100: | Improved |
| Use LogME instead of the strongest prior ranking method | Pre-trained model selection · Birdsnap and Pets | weighted Kendall tau: Birdsnap: 0.67 vs NCE 0.74; Pets: 0.58 vs NCE 0.84 | Hurt |
| Use LogME to rank contrastive pre-trained models | Contrastive pre-trained models · Aircraft classification and dSprites regression | weighted Kendall tau: Aircraft: 1.0; dSprites: 1.0 | Improved |
| Use LogME to rank pre-trained language models | Pre-trained language models · GLUE tasks | weighted Kendall tau: MNLI: 0.66; QQP: 0.73; QNLI: 1.00; SST-2: 0.68; CoLA: 1.00; MRPC: 0.53; RTE: 1.00 | Mixed |
| Replace maximum evidence with the best-fit likelihood (ablation) | LogME feature-label compatibility assessment · Supplementary B experiments | assessment quality | Hurt |
| Retrain a downstream classification head instead of computing LogME (ablation) | Pre-trained model selection on Caltech · Caltech | correlation with transfer performance: 3× less time for LogME than re-training a head with one fixed hyper-parameter | Hurt |
Later papers that compared with this method
14 later accepted papers report LogME as a baseline in their results tables; 12 of them report a better number on most of the results they share with it.
- How NOT to benchmark your SITE metric: Beyond Static Leaderboards and Towards Realistic Evaluation. (ICLR 2026 · better on 7 of 7 shared results)
- ModelLens: Finding the Best for Your Task from Myriads of Models (NeurIPS 2026 · better on 1 of 1 shared result)
- RADAR: Relative Angular Divergence Across Representations (NeurIPS 2026 · better on 0 of 1 shared result)
- Know2Vec: A Black-Box Proxy for Neural Network Retrieval (AAAI 2025 · better on 4 of 4 shared results)
- Selection, Ensemble, and Adaptation: Advancing Multi-Source-Free Domain Adaptation via Architecture Zoo (TPAMI 2025 · better on 1 of 1 shared result)
- Evaluating Transferability in Retrieval Tasks: An Approach Using MMD and Kernel Methods (CVPR 2024 · better on 7 of 7 shared results)
- LEAD: Exploring Logit Space Evolution for Model Selection (CVPR 2024 · better on 1 of 1 shared result)
- How Far Pre-trained Models Are from Neural Collapse on the Target Dataset Informs their Transferability (ICCV 2023 · better on 2 of 2 shared results)
6 more not listed.
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.
- LEEP — LEEP: A New Measure to Evaluate Transferability of Learned Representations (ICML 2020) · better on 4 of 4 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
- Assessing Pre-Trained Models for Transfer Learning Through Distribution of Spectral Components (AAAI 2025 · similarity 0.86)
- LEAD: Exploring Logit Space Evolution for Model Selection (CVPR 2024 · similarity 0.85)
- Transfer Learning of Linear Regression with Multiple Pretrained Models: Benefiting from More Pretrained Models via Overparameterization Debiasing (NeurIPS 2026 · similarity 0.85)
- Understanding the Transferability of Representations via Task-Relatedness (NeurIPS 2024 · similarity 0.85)
- Transferring Knowledge From Large Foundation Models to Small Downstream Models (ICML 2024 · similarity 0.85)
- A Bayesian Model Selection Criterion for Selecting Pretraining Checkpoints (ICML 2025 · 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.
