Research map / Learning theory and algorithms / Prediction and decision losses
Generalized Coverage for More Robust Low-Budget Active Learning
Wonho Bae, Junhyug Noh, Danica J. Sutherland
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 |
|---|---|---|---|
| use smooth Gaussian kernel generalized coverage instead of ProbCover top-hat with radius delta (ablation) | ProbCover · CIFAR10 (SimCLR features, 1-NN) | test accuracy | Improved |
| add MaxHerding selection instead of Typiclust | low-budget active learning · ImageNet (DINO features) | test accuracy | Improved |
| use uncertainty-based selection (Entropy, Margin) instead of random in low budget | active learning · CIFAR benchmarks low-budget | test accuracy | Hurt |
| use non-greedy kernel k-medoids instead of greedy MaxHerding (ablation) | generalized coverage maximization · CIFAR100 | coverage and test accuracy: more than ten times the runtime | No clear effect |
| swap Gaussian kernel for Student-t, Laplace or Cauchy kernels (ablation) | MaxHerding · CIFAR10 and CIFAR100 | test accuracy | No clear effect |
| use MaxHerding in high-budget regime (1,000 samples per iteration) (ablation) | MaxHerding · CIFAR10 high budget | test accuracy | Mixed |
| apply MaxHerding to object detection selection instead of Typiclust or Coreset | YOLOv5 on VOC · PASCAL VOC (YOLOv5) | mAP50-95: 10.9 -> 13.4 at |L|=400 (Typiclust vs MaxHerding) | Improved |
| use MaxHerding versus ProbCover on imbalanced data | low-budget active learning · Imbalanced CIFAR100 and TinyImageNet | test accuracy | Improved |
Closest papers by meaning
- Active Learning Through a Covering Lens (NeurIPS 2022 · similarity 0.90)
- Uncertainty Herding: One Active Learning Method for All Label Budgets (ICLR 2025 · similarity 0.87)
- Active Covering (ICML 2021 · similarity 0.87)
- Online Active Learning with Surrogate Loss Functions (NeurIPS 2021 · similarity 0.86)
- Low-Budget Active Learning via Wasserstein Distance: An Integer Programming Approach (ICLR 2022 · similarity 0.86)
- 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.
