Research map / Learning theory and algorithms / Prediction and decision losses
Margin-based sampling in high dimensions: When being active is less efficient than staying passive
Alexandru Tifrea, Jacob Clarysse, Fanny Yang
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 margin-based active sampling instead of uniform sampling | Logistic regression · High-dimensional datasets with low query budgets | test error | Hurt |
| Use Bayes-optimal margins for query selection instead of uniform sampling | Logistic regression · High-dimensional, noiseless classification | population error | Hurt |
| Increase class separation by removing the 25% or 50% closest samples to the decision boundary (ablation) | Margin-based active learning · High-dimensional real-world datasets | test error | Improved |
| Increase the seed set size (ablation) | Margin-based active learning · Synthetic and real-world high-dimensional data | test error | Improved |
| Use oracle margins instead of margins from an empirical classifier (ablation) | Margin-based active learning · High-dimensional datasets | test error | Hurt |
| Combine margin-based sampling with uniform sampling using an epsilon-greedy strategy (ablation) | Epsilon-greedy active learning · High-dimensional datasets | test error | Mixed |
| Use coreset-based sampling instead of margin-based sampling (ablation) | Coreset-based active learning · Real-world high-dimensional datasets | test error | Improved |
| Use coreset-based sampling instead of passive uniform sampling (ablation) | Coreset-based active learning · Real-world high-dimensional datasets | test error | Hurt |
| Exclude points closest to the empirical decision boundary from query candidates (ablation) | Margin-based active learning · High-dimensional datasets | test error | No clear effect |
| Increase the labeling budget (ablation) | Margin-based active learning · High-dimensional datasets | test error | Improved |
Closest papers by meaning
- Regimes of No Gain in Multi-class Active Learning (JMLR 2024 · similarity 0.89)
- Constants Matter: The Performance Gains of Active Learning (ICML 2022 · similarity 0.88)
- Discover-Then-Rank Unlabeled Support Vectors in the Dual Space for Multi-Class Active Learning (ICML 2023 · similarity 0.88)
- Margin-Based Active Learning of Classifiers (JMLR 2024 · similarity 0.87)
- A Cross-Domain Benchmark for Active Learning (NeurIPS 2024 · similarity 0.85)
- Improved Active Learning via Dependent Leverage Score Sampling (ICLR 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.
