Research map / Label-efficient and robust learning / Recognition and long-tail
How to Select Which Active Learning Strategy is Best Suited for Your Specific Problem and Budget
Guy Hacohen, Daphna Weinshall
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 |
|---|---|---|---|
| dynamically choose low-, random-, or high-budget query strategies using the removal-based derivative test | active learning with ResNet-18 · CIFAR-10, CIFAR-100, and ImageNet-50 across low, medium, and high budgets | accuracy | Improved |
| use SelectAL rather than the best single query strategy | active learning · CIFAR-10, 100+100; CIFAR-100, 100+100; ImageNet-50, 100+100 | accuracy: 35.1 ± 0.3; 8.2 ± 0.1; 11.4 ± 0.6 | No clear effect |
| use SelectAL rather than the best single query strategy | active learning · CIFAR-10, 25k+5k; CIFAR-100, 30k+7k; ImageNet-50, 25k+5k | accuracy: 88.1 ± 0.1; 61.9 ± 0.1; 80.8 ± 0.2 | No clear effect |
| choose strategies dynamically instead of using ProbCover alone | active learning · CIFAR-10 and CIFAR-100, high budgets | accuracy: 87.1 ± 0.1 -> 88.1 ± 0.1; 61.4 ± 0.3 -> 61.9 ± 0.1 | Improved |
| use BADGE as the high-budget removal-test strategy instead of inverse-TypiClust (ablation) | SelectAL's budget-regime test · CIFAR-10 | accuracy-based transition detection | Hurt |
| allow the regime-test strategy to use the labeled set (ablation) | SelectAL's budget-regime test · CIFAR-10 | accuracy-based transition detection | Hurt |
| use MoCo, SimCLR, or SCAN features in the removal test (ablation) | SelectAL · CIFAR-10 | accuracy-based strategy selection | No clear effect |
| replace the pixel representation with MoCo, DINO, SimCLR, or SWAV features (ablation) | TypiClust · TinyImageNet, low-budget regime | test accuracy | Improved |
| use an autoencoder representation instead of the other tested feature spaces (ablation) | TypiClust · TinyImageNet, low-budget regime | test accuracy | Hurt |
| use ProbCover for low budgets and BADGE for high budgets rather than a single strategy across budgets | active learning · CIFAR-10, CIFAR-100, and ImageNet-50 | accuracy | Mixed |
Closest papers by meaning
- Active Learning on a Budget: Opposite Strategies Suit High and Low Budgets (ICML 2022 · similarity 0.87)
- Contextual Active Model Selection (NeurIPS 2024 · similarity 0.83)
- AutoAL: Automated Active Learning with Differentiable Query Strategy Search (ICML 2025 · similarity 0.83)
- A Cross-Domain Benchmark for Active Learning (NeurIPS 2024 · similarity 0.83)
- Provably Neural Active Learning Succeeds via Prioritizing Perplexing Samples (ICML 2024 · similarity 0.82)
- Composite Active Learning: Towards Multi-Domain Active Learning with Theoretical Guarantees (AAAI 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.
