Research map / Label-efficient and robust learning / Recognition and long-tail
SIMILAR: Submodular Information Measures Based Active Learning In Realistic Scenarios
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 |
|---|---|---|---|
| Use rare-class query relevance with SMI acquisition (Flqmi or Logdetmi) instead of existing active-learning baselines | SIMILAR · CIFAR-10, MNIST, and ImageNet rare-class setting | rare-class accuracy: ≈ 10-18% | Helped |
| Use rare-class query relevance with SMI acquisition (Flqmi or Logdetmi) instead of existing active-learning baselines | SIMILAR · CIFAR-10, MNIST, and ImageNet rare-class setting | overall accuracy: ≈ 5-10% | Helped |
| Condition selection on the already labeled set with SCG | SIMILAR · CIFAR-10 and MNIST with redundant unlabeled data | accuracy: ≈ 3-5% | Helped |
| Condition selection on labeled OOD examples and target relevance to labeled ID examples with SCMI | SIMILAR · CIFAR-10 and MNIST with OOD unlabeled data | ID accuracy: ≈ 5 - 10% | Helped |
| Add conditioning on labeled OOD examples rather than using SMI alone (ablation) | SIMILAR · CIFAR-10 and MNIST with OOD unlabeled data | ID accuracy: 2-3% | Helped |
| Use Flqmi or Logdetmi, which balance diversity and query relevance, rather than Gcmi, which models relevance only (ablation) | SIMILAR rare-class acquisition · Rare-class setting | accuracy | Helped |
| Combine Gcmi with a linear diversity term (Div-Gcmi) instead of using the richer SMI acquisition (ablation) | SIMILAR rare-class acquisition · Rare-class setting | accuracy | Hurt |
| Use Fisher with a 40× larger validation set instead of the small realistic validation set (ablation) | Fisher baseline · Rare-class setting | accuracy: 40 × | Helped |
| Use a larger validation set for Fisher rather than the small realistic validation set (ablation) | Fisher baseline · Rare-class setting | accuracy | Helped |
| Use Logdetmi rather than other methods under extreme class imbalance (ablation) | SIMILAR rare-class acquisition · MNIST, ρ = 100, B = 25 | accuracy: 100x | Helped |
| Condition batch selection on the updated labeled set to reduce inter-batch redundancy (ablation) | SIMILAR · Redundant unlabeled data | number of unique points selected | Helped |
| Use the facility-location function in standard active learning rather than Logdet or Badge (ablation) | SIMILAR · Standard active learning | accuracy | Hurt |
Closest papers by meaning
- Active Learning by Feature Mixing (CVPR 2022 · similarity 0.87)
- Composite Active Learning: Towards Multi-Domain Active Learning with Theoretical Guarantees (AAAI 2024 · similarity 0.87)
- A Cross-Domain Benchmark for Active Learning (NeurIPS 2024 · similarity 0.86)
- Towards Cost-Effective Learning: A Synergy of Semi-Supervised and Active Learning (CVPR 2025 · similarity 0.85)
- Plug and Play Active Learning for Object Detection (CVPR 2024 · similarity 0.85)
- Fast Co-Training under Weak Dependence via Stream-Based Active Learning (ICML 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.