Research map / Label-efficient and robust learning / Recognition and long-tail
Active Finetuning: Exploiting Annotation Budget in the Pretraining-Finetuning Paradigm
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 |
|---|---|---|---|
| Select a representative, diverse subset by continuous feature-space optimization | ActiveFT · CIFAR10 1% | Top-1 accuracy: 83.4±0.3 -> 88.2±0.4 | Helped |
| Select a representative, diverse subset by continuous feature-space optimization | ActiveFT · CIFAR100 2% | Top-1 accuracy: 35.3±0.8 -> 40.7±0.9 | Helped |
| Select a representative, diverse subset by continuous feature-space optimization | ActiveFT · ImageNet-1k 1% | Top-1 accuracy: 45.1±0.8 -> 50.1±0.3 | Helped |
| Select a representative, diverse subset by continuous feature-space optimization | ActiveFT · ADE20k 10% | mIoU: 20.27 -> 21.60±0.40 | Helped |
| Update each sample's closest parameter assignment c_i during optimization (ablation) | ActiveFT · CIFAR100 2% | Top-1 accuracy: 20.6 -> 40.7 | Helped |
| Update each sample's closest parameter assignment c_i during optimization (ablation) | ActiveFT · CIFAR100 5% | Top-1 accuracy: 52.8 -> 54.6 | Helped |
| Add the proposed diversity regularization rather than removing it (ablation) | ActiveFT · CIFAR100 2% | Top-1 accuracy: 33.1 -> 40.7 | Helped |
| Use the proposed diversity regularizer instead of InfoNCE (ablation) | ActiveFT · CIFAR100 2% | Top-1 accuracy: 26.8 -> 40.7 | Helped |
| Use the proposed diversity regularizer instead of InfoNCE (ablation) | ActiveFT · CIFAR100 5% | Top-1 accuracy: 46.9 -> 54.6 | Helped |
| Set temperature to τ=0.5 instead of τ=0.07 (ablation) | ActiveFT · CIFAR10 1% | Top-1 accuracy: 88.2 -> 86.1 | Hurt |
| Optimize the parametric model for 100 iterations rather than 50 (ablation) | ActiveFT · ImageNet-1k 1% | Top-1 accuracy: 46.7 -> 50.2 | Helped |
| Increase optimization from 100 to 200 iterations (ablation) | ActiveFT · ImageNet-1k 1% | Top-1 accuracy: 50.2 -> 50.1 | No clear effect |
Closest papers by meaning
- ActiveDC: Distribution Calibration for Active Finetuning (CVPR 2024 · similarity 0.90)
- Boundary Matters: A Bi-Level Active Finetuning Method (NeurIPS 2024 · similarity 0.90)
- A Cross-Domain Benchmark for Active Learning (NeurIPS 2024 · similarity 0.87)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.86)
- Pixels to Tokens: Token Space Efficient Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2026 · similarity 0.84)
- Density Matters: Improved Core-Set for Active Domain Adaptive Segmentation (AAAI 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.