Research map / Optimization / Optimizers
Minimal Variance Sampling in Stochastic Gradient Boosting
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 MVS sampling instead of the full-data baseline | CatBoost · 153 binary classification datasets | mean ROC-AUC: +0.12% mean ROC-AUC improvement | Helped |
| use MVS rather than SGB or GOSS sampling | LightGBM · seven classification datasets, average | relative error change: MVS -0.74%; SGB -0.22%; GOSS -0.28% | Helped |
| use MVS sampling | LightGBM · KDD Internet | relative error change: -3.03% | Helped |
| use MVS sampling | LightGBM · KDD Churn | relative error change: +0.17% | Hurt |
| replace tuned lambda with squared mean of the initial leaf (MVS Adaptive) (ablation) | MVS · seven classification datasets, average | relative error change: MVS Adaptive -0.70%; MVS -0.74% | Helped |
| replace tuned lambda with squared mean of the initial leaf (MVS Adaptive) (ablation) | MVS · KDD Upselling | relative error change: -0.07% -> -0.28% | Helped |
| use MVS rather than SGB or GOSS at small sample rates | LightGBM · average over datasets, sample rate 0.1 | relative error change: SGB +6.83%; GOSS +8.00%; MVS +3.69% | Helped |
| use MVS rather than GOSS to reach baseline quality on Higgs | LightGBM · Higgs | relative learning time change: -17.7% versus -8.5% | Helped |
| use MVS sampling rather than SGB or GOSS | LightGBM · datasets in this paper | relative learning time change: MVS -27.7%; SGB -20.7%; GOSS -20.4% | Helped |
| set lambda adaptively from the initial leaf instead of tuning it (ablation) | MVS · sampling parameter tuning | tuning time | Helped |
Closest papers by meaning
- SGLB: Stochastic Gradient Langevin Boosting (ICML 2021 · similarity 0.87)
- Adaptive Sampling for SGD by Exploiting Side Information (ICML 2016 · similarity 0.87)
- Stochastic Multiple Target Sampling Gradient Descent (NeurIPS 2022 · similarity 0.87)
- Statistical Inference for Gradient Boosting Regression (NeurIPS 2025 · similarity 0.86)
- Robust Minimax Boosting with Performance Guarantees (NeurIPS 2025 · similarity 0.85)
- Supervised Score-Based Modeling by Gradient Boosting (AAAI 2025 · 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.