Research map / Label-efficient and robust learning / Recognition and long-tail
Boosting Active Learning via Improving Test Performance
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 annotation data by expected gradient norm | Active learning with ResNet-18-selected data used to train ResNet-50 · Cifar10, 20,000 selected samples | accuracy: 94.1 | Helped |
| Select annotation data by entropy gradient norm | Active learning with ResNet-18-selected data used to train ResNet-50 · Cifar10, 20,000 selected samples | accuracy: 93.16 | Helped |
| Select data by gradient norm rather than training with all training data | Cifar10 active learning · Cifar10, final AL cycle versus full-data training | accuracy: 93.16% -> 94.39% | Helped |
| Use expected empirical loss to estimate unlabeled-sample gradient norm | Active learning · Cifar10, ResNet-50 trained on selected data | accuracy: 94.1 | Helped |
| Use entropy loss to estimate unlabeled-sample gradient norm | Active learning · Cifar10 and Cityscapes | test performance | Helped |
| Use gradient-norm-selected data to train a deeper model from scratch (ablation) | ResNet-50 trained using data selected by ResNet-18 · Cifar10 | accuracy: 94.1 | Helped |
| Use entropy-gradnorm rather than full-data training (ablation) | ResNet-18 · Cifar10 | accuracy: 93.16% -> 94.39% | Helped |
| Estimate gradient norm using expected loss rather than entropy loss (ablation) | Active learning · Cifar10, ResNet-50 trained on selected data | accuracy: 94.1 vs 93.16 | Mixed |
Closest papers by meaning
- A Cross-Domain Benchmark for Active Learning (NeurIPS 2024 · similarity 0.87)
- Prediction-Powered Active Testing (NeurIPS 2026 · similarity 0.86)
- Rethinking Epistemic and Aleatoric Uncertainty for Active Open-Set Annotation: An Energy-Based Approach (CVPR 2025 · similarity 0.86)
- Instance-wise Supervision-level Optimization in Active Learning (CVPR 2025 · similarity 0.86)
- Towards Cost-Effective Learning: A Synergy of Semi-Supervised and Active Learning (CVPR 2025 · similarity 0.85)
- Universal Rates for Active Learning (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.