Research map / Label-efficient and robust learning / Recognition and long-tail
Active Feature Selection for the Mutual Information Criterion
Shachar Schnapp, Sivan Sabato
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 AFS to actively choose examples to label rather than RANDOM, CORESET, or DWUS | Mutual-information feature selection · Benchmark datasets, including MUSK, MNIST, and RELATHE | mutual information gap | Improved |
| Use the entropy-targeted I-CP allocation instead of the other single-feature allocation strategies | Single-feature conditional entropy estimation · Synthetic and Adult experiments | absolute entropy estimation error | Improved |
| Use single-feature allocation scores rather than selecting examples by one random feature (ablation) | AFS · Feature-selection datasets, k in {5,10,20} | mutual information gap | Mixed |
| Add the pairwise sampling-bias correction term (ablation) | AFS · Feature-selection datasets, k in {5,10,20} | mutual information gap | Improved |
| Choose examples using the selected feature subset F^(t) rather than averaging scores over all features (ablation) | AFS · Feature-selection datasets, k in {5,10,20} | mutual information gap | Improved |
| Add the safeguard that switches remaining labels to random selection when the top-k estimate stops changing (ablation) | AFS · Feature-selection datasets | mutual information gap | Improved |
| Replace the aggregation function ℓ1 with ℓ2 (ablation) | AFS · Feature-selection datasets | mutual information gap | No clear effect |
| Replace the aggregation function ℓ1 with ℓ∞ (ablation) | AFS · Feature-selection datasets | mutual information gap | Hurt |
Closest papers by meaning
- Estimating Conditional Mutual Information for Dynamic Feature Selection (ICLR 2024 · similarity 0.87)
- Stochastic Encodings for Active Feature Acquisition (ICML 2025 · similarity 0.85)
- Towards a statistical theory of data selection under weak supervision (ICLR 2024 · similarity 0.85)
- SAND: One-Shot Feature Selection with Additive Noise Distortion (ICML 2025 · similarity 0.84)
- Active feature acquisition via explainability-driven ranking (ICML 2025 · similarity 0.84)
- Best Subset Selection: Optimal Pursuit for Feature Selection and Elimination (ICML 2025 · 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.
