Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Maximizing Overall Diversity for Improved Uncertainty Estimates in Deep Ensembles
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 |
|---|---|---|---|
| Add a diversity penalty measured on uniformly sampled inputs across the feature space | DeepEns · Protein-DNA binding, top 10% affinity OOD | OOD NLL: 0.7485$\pm$0.124 -> 0.7153$\pm$0.117 | Helped |
| Add a diversity penalty measured on uniformly sampled inputs across the feature space | DeepEns · Protein-DNA binding, >80% GC content OOD | OOD NLL: -0.6938$\pm$0.052 -> -0.7007$\pm$0.047 | Mixed |
| Reweight augmented examples by their distance from the training data (ablation) | MOD · Protein-DNA binding, top 10% affinity OOD | OOD NLL: 0.7153$\pm$0.117 -> 0.7225$\pm$0.116 | Hurt |
| Measure diversity only on training examples instead of across the input space (ablation) | MOD · Protein-DNA binding, top 10% affinity OOD | OOD NLL: 0.7153$\pm$0.117 -> 0.7326$\pm$0.121 | Hurt |
| Use a diversity penalty to train the ensemble | DeepEns · Protein-DNA binding Bayesian optimization | simple regret: 26 (0.003) | Helped |
| Use adversarial optimization to target relatively over-confident OOD points (ablation) | MOD · IMDB-Wiki age prediction, oldest 10% OOD | OOD NLL: 1.1294$\pm$0.1707 -> 1.1547$\pm$0.1865 | Hurt |
| Increase the diversity penalty beyond a suitable value (ablation) | MOD · UCI and protein-binding regression tasks | in-distribution NLL | Hurt |
Closest papers by meaning
- DIBS: Diversity Inducing Information Bottleneck in Model Ensembles (AAAI 2021 · similarity 0.91)
- Neural Ensemble Search for Uncertainty Estimation and Dataset Shift (NeurIPS 2021 · similarity 0.88)
- Improving Adversarial Robustness via Promoting Ensemble Diversity (ICML 2019 · similarity 0.87)
- Structure-informed Risk Minimization for Robust Ensemble Learning (ICML 2025 · similarity 0.86)
- On the Diversity of Adversarial Ensemble Learning (ICML 2025 · similarity 0.85)
- Density-Softmax: Efficient Test-time Model for Uncertainty Estimation and Robustness under Distribution Shifts (ICML 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.