Research map / Learning theory and algorithms / Prediction and decision losses
When are ensembles really effective?
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 an ensemble when disagreement is large relative to average error (DER > 1) | Competent majority-vote classifiers · K-class classification | EIR | Helped |
| Use an ensemble when disagreement is small relative to average error (DER < 1) | Competent majority-vote classifiers · K-class classification | EIR | No clear effect |
| Ensemble competent classifiers rather than using a single classifier | Competent ensembles · K-class classification | error rate | Helped |
| Increase model capacity past the interpolation threshold | Bagged random feature classifiers · MNIST, Thyroid, and QSAR | EIR | Hurt |
| Increase model capacity past the interpolation threshold | ResNet18 ensembles · CIFAR-10 | EIR | Hurt |
| Fine-tune large BERT models and ensemble them | BERT ensembles · GLUE classification tasks | EIR | No clear effect |
| Ensemble non-interpolating Bayesian neural network samples | Bayesian neural networks · CIFAR-10 and IMDB | EIR | Helped |
| Increase tree capacity through the interpolation threshold | Random forests · MNIST, Thyroid, and QSAR | EIR | Helped |
| Increase tree capacity beyond the interpolation threshold | Random forests · MNIST, Thyroid, and QSAR | EIR | No clear effect |
| Compare EIR with DER across varied model capacities | Random feature, random forest, and neural-network ensembles · MNIST, CIFAR-10, QSAR, and Thyroid | Pearson correlation: Pearson R ≥ 0.96 | Helped |
| Use the proposed disagreement-based majority-vote bound instead of the prior binary disagreement bound | Majority-vote error bounds · Binary classification | bound tightness: factor of 2 | Helped |
Closest papers by meaning
- Subsampled Ensemble Can Improve Generalization Tail Exponentially (NeurIPS 2025 · similarity 0.87)
- How many classifiers do we need? (NeurIPS 2024 · similarity 0.87)
- Exact power indices for plurality-voting ensembles (NeurIPS 2026 · similarity 0.86)
- Rethinking LLM Ensembling from the Perspective of Mixture Models (ICML 2026 · similarity 0.84)
- Ex Uno Pluria: Insights on Ensembling in Low Precision Number Systems (NeurIPS 2024 · similarity 0.84)
- Don’t Label Twice: Quantity Beats Quality when Comparing Binary Classifiers on a Budget (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.