Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
MESA: Boost Ensemble Imbalanced Learning with MEta-SAmpler
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 the learned adaptive meta-sampler instead of the strongest listed under-sampling ensemble baseline | Decision-tree ensemble imbalanced learning · Optical Digits, Spectrometer, ISOLET, and Mammography | AUCPRC: 0.959 -> 0.968; 0.784 -> 0.803; 0.841 -> 0.877; 0.575 -> 0.644 | Helped |
| replace the learned sampler policy with a randomly initialized policy (ablation) | MESA · Optical Digits, k=5 | AUCPRC: 0.929 -> 0.904 | Hurt |
| replace meta-sampling with random sampling (ablation) | MESA · Optical Digits, k=5 | AUCPRC: 0.929 -> 0.876 | Hurt |
| replace the learned sampler policy with a randomly initialized policy (ablation) | MESA · Spectrometer, k=5 | AUCPRC: 0.723 -> 0.685 | Hurt |
| replace meta-sampling with random sampling (ablation) | MESA · Spectrometer, k=5 | AUCPRC: 0.723 -> 0.610 | Hurt |
| replace the learned sampler policy with a randomly initialized policy (ablation) | MESA · ISOLET, k=5 | AUCPRC: 0.787 -> 0.748 | Hurt |
| replace meta-sampling with random sampling (ablation) | MESA · ISOLET, k=5 | AUCPRC: 0.787 -> 0.688 | Hurt |
| replace the learned sampler policy with a randomly initialized policy (ablation) | MESA · Mammography, k=5 | AUCPRC: 0.515 -> 0.405 | Hurt |
| replace meta-sampling with random sampling (ablation) | MESA · Mammography, k=5 | AUCPRC: 0.515 -> 0.307 | Hurt |
| vary the policy-network architecture and width (ablation) | MESA · Optical Digits | AUCPRC: 0.929±0.015 -> 0.930±0.014 | No clear effect |
| train the meta-sampler on 10% rather than 100% of the Protein Homo. task (ablation) | MESA · Protein Homo., k=10 | AUCPRC: 0.840±0.009 -> 0.839±0.008; -0.10% | No clear effect |
| train the meta-sampler on Spectrometer and transfer it to Mammography (ablation) | MESA · Mammography, k=10 | AUCPRC: 0.644 -> 0.641; -0.54% | No clear effect |
Closest papers by meaning
- Learning under Localized Minority Imbalance (NeurIPS 2026 · similarity 0.85)
- Sharpness-diversity tradeoff: improving flat ensembles with SharpBalance (NeurIPS 2024 · similarity 0.84)
- Concentration and excess risk bounds for imbalanced classification with synthetic oversampling (NeurIPS 2025 · similarity 0.84)
- Restoring balance: principled under/oversampling of data for optimal classification (ICML 2024 · similarity 0.84)
- Improving Generalization via Meta-Learning on Hard Samples (CVPR 2024 · similarity 0.84)
- Class-Attribute Priors: Adapting Optimization to Heterogeneity and Fairness Objective (AAAI 2024 · similarity 0.83)
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.