Research map / Label-efficient and robust learning / Recognition and long-tail
LEEP: A New Measure to Evaluate Transferability of Learned Representations
Cuong Nguyen, Tal Hassner, Matthias Seeger, Cedric Archambeau
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 LEEP instead of NCE to estimate transfer performance | Re-train head, CIFAR100 large balanced from ImageNet · CIFAR100, large balanced | Pearson correlation with test accuracy: 0.974 vs 0.973 | No clear effect |
| Use LEEP instead of NCE to estimate transfer performance | Re-train head, CIFAR100 small balanced from ImageNet · CIFAR100, small balanced | Pearson correlation with test accuracy: 0.715 -> 0.798 | Improved |
| Use LEEP instead of NCE to estimate transfer performance | Re-train head, CIFAR100 small balanced noisy from ImageNet · CIFAR100, small balanced noisy | Pearson correlation with test accuracy: 0.579 -> 0.612 | Improved |
| Use LEEP instead of NCE to estimate transfer performance | Re-train head, CIFAR100 small imbalanced from ImageNet · CIFAR100, small imbalanced | Pearson correlation with test F1: 0.484 -> 0.522 | Improved |
| Use LEEP instead of H score to estimate transfer performance | Re-train head, CIFAR100 small balanced from ImageNet · CIFAR100, small balanced | Pearson correlation with test accuracy: 0.026* -> 0.798 | Improved |
| Use LEEP instead of H score to estimate transfer performance | Fine-tune, CIFAR100 small balanced from ImageNet · CIFAR100, small balanced | Pearson correlation with test accuracy: -0.029* -> 0.762 | Improved |
| Use LEEP instead of NCE to estimate meta-transfer performance | CNAPS, CIFAR100 small balanced from ImageNet · CIFAR100, small balanced | Pearson correlation with test accuracy: 0.310 -> 0.591 | Improved |
| Use LEEP instead of NCE to estimate transfer performance | Fine-tune, CIFAR100 small balanced from CIFAR10 · CIFAR100, small balanced | Pearson correlation with test accuracy: 0.401 -> 0.396 | Hurt |
| Use LEEP instead of NCE to estimate transfer performance | Fine-tune, CIFAR100 large balanced from ImageNet · CIFAR100, large balanced | Pearson correlation with test accuracy: 0.945 -> 0.944 | Hurt |
| Use LEEP instead of H score to estimate transfer performance | Re-train head, FashionMNIST small balanced from CIFAR10 · FashionMNIST, small balanced | Pearson correlation with test accuracy: 0.787 -> 0.518 | Hurt |
| Use LEEP instead of H score to estimate transfer performance | Fine-tune, FashionMNIST small balanced from CIFAR10 · FashionMNIST, small balanced | Pearson correlation with test accuracy: 0.826 -> 0.339 | Hurt |
| Use LEEP instead of H score to estimate transfer performance | Fine-tune, CIFAR100 small imbalanced from CIFAR10 · CIFAR100, small imbalanced | Pearson correlation with test F1: 0.758 -> 0.597 | Hurt |
Later papers that compared with this method
12 later accepted papers report LEEP as a baseline in their results tables; 11 of them report a better number on most of the results they share with it.
- RADAR: Relative Angular Divergence Across Representations (NeurIPS 2026 · better on 4 of 4 shared results)
- Unsupervised Source-Free Ranking of Biomedical Segmentation Models Under Distribution Shift (ECCV 2026 · better on 1 of 2 shared results)
- Know2Vec: A Black-Box Proxy for Neural Network Retrieval (AAAI 2025 · better on 2 of 2 shared results)
- Selection, Ensemble, and Adaptation: Advancing Multi-Source-Free Domain Adaptation via Architecture Zoo (TPAMI 2025 · 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 1 of 1 shared result)
- Pre-Trained Model Reusability Evaluation for Small-Data Transfer Learning (NeurIPS 2022 · better on 12 of 12 shared results)
- How Stable Are Transferability Metrics Evaluations? (ECCV 2022 · better on 4 of 4 shared results)
- Transferability Estimation Using Bhattacharyya Class Separability (CVPR 2022 · better on 2 of 3 shared results)
4 more not listed.
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
- Revisiting the Transferability of Supervised Pretraining: An MLP Perspective (CVPR 2022 · similarity 0.86)
- A Unified Few-Shot Classification Benchmark to Compare Transfer and Meta Learning Approaches (NeurIPS 2021 · similarity 0.84)
- How Classifier Features Transfer to Downstream: An Asymptotic Analysis in a Two-Layer Model (NeurIPS 2025 · similarity 0.84)
- Fast and Accurate Transferability Measurement by Evaluating Intra-class Feature Variance (ICCV 2023 · similarity 0.84)
- A Mathematical Framework for Quantifying Transferability in Multi-source Transfer Learning (NeurIPS 2021 · similarity 0.83)
- Universality in Transfer Learning for Linear Models (NeurIPS 2024 · similarity 0.82)
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.
