Research map / Label-efficient and robust learning / Recognition and long-tail
A Data-Based Perspective on Transfer Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Remove the most negatively influential ImageNet classes identified for the target task | ImageNet-pretrained ResNet-18 transferred to CIFAR-10 · CIFAR-10 | test accuracy: 2.5\% larger than what using the entire ImageNet dataset would yield | Helped |
| Remove the most negatively influential source classes, selected using task-specific influences | ImageNet-pretrained models · 11 target classification tasks | transfer accuracy | Helped |
| Remove the most positively influential ImageNet classes | ImageNet-pretrained model transferred to CIFAR-10 · CIFAR-10 | test accuracy | Hurt |
| Remove the most negatively influential ImageNet class for an individual target example | ImageNet-pretrained model transferred to CIFAR-10 · Highly negatively influenced CIFAR-10 examples | prediction correctness: 61.5\% were predicted correctly more often; 34.6\% were predicted incorrectly more often; 3.9\% were predicted correctly | Mixed |
| Use alternative targets such as correctness or margins to compute influences rather than the ground-truth softmax output (ablation) | Influence-based class ranking for ImageNet → CIFAR-10 · CIFAR-10 | test accuracy | No clear effect |
| Use datamodels instead of influences to rank ImageNet classes for removal (ablation) | Class-based counterfactual transfer experiment · CIFAR-10 | test accuracy | No clear effect |
| Use CIFAR-10 influence estimates to choose ImageNet classes for removal on other target datasets (ablation) | ImageNet-pretrained models · Other target classification datasets | transfer accuracy | Mixed |
| Apply ResNet-18-derived influence rankings when training a ResNet-50 (ablation) | ImageNet-pretrained ResNet-50 · ImageNet transfer tasks | transfer accuracy | Helped |
Closest papers by meaning
- What is being transferred in transfer learning? (NeurIPS 2020 · similarity 0.89)
- Do Better ImageNet Models Transfer Better? (CVPR 2019 · similarity 0.88)
- Selectivity Drives Productivity: Efficient Dataset Pruning for Enhanced Transfer Learning (NeurIPS 2023 · similarity 0.87)
- A High-Dimensional Statistical Method for Optimizing Transfer Quantities in Multi-Source Transfer Learning (NeurIPS 2025 · similarity 0.86)
- Features are fate: a theory of transfer learning in high-dimensional regression (ICML 2025 · similarity 0.85)
- Transfer Learning with Informative Priors: Simple Baselines Better than Previously Reported (NeurIPS 2024 · similarity 0.85)
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.