Research map / Label-efficient and robust learning / Recognition and long-tail
Combining Ensembles and Data Augmentation Can Harm Your Calibration
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 Mixup to BatchEnsemble | Wide ResNet-28-10 BatchEnsemble · CIFAR-10 | ECE: 1.8 ± 0.2 % -> 6.4 ± 0.4 % | Hurt |
| Add Mixup to BatchEnsemble | Wide ResNet-28-10 BatchEnsemble · CIFAR-10 | accuracy: 96.22 ± 0.07 -> 96.98±0.08 | Helped |
| Apply class-adjusted Mixup only to classes with over-confidence (ablation) | BatchEnsemble · CIFAR-10 | ECE: 6.4 ± 0.4 % -> 1.2 ± 0.2 % | Helped |
| Apply class-adjusted Mixup only to classes with over-confidence (ablation) | BatchEnsemble · CIFAR-10 | accuracy: 96.98±0.08 -> 96.94 ± 0.10 | No clear effect |
| Add Mixup to BatchEnsemble | Wide ResNet-28-10 BatchEnsemble · CIFAR-100 | ECE: 2.8±0.1% -> 9.7±0.5 % | Hurt |
| Apply class-adjusted Mixup only to classes with over-confidence (ablation) | BatchEnsemble · CIFAR-100 | ECE: 9.7±0.5 % -> 2.3±0.1% | Helped |
| Add Mixup to BatchEnsemble | ResNet-50 BatchEnsemble · ImageNet | ECE: 2.0% -> 2.1% | Hurt |
| Apply class-adjusted Mixup instead of uniform Mixup (ablation) | ResNet-50 BatchEnsemble · ImageNet | ECE: 2.1% -> 1.5% | Helped |
| Add Mixup to AugMix BatchEnsemble | Wide ResNet-28-10 BatchEnsemble · CIFAR-10 | ECE: 1.02% -> 1.71% | Hurt |
| Replace uniform Mixup with class-adjusted Mixup in AugMix (ablation) | BatchEnsemble · CIFAR-10 | ECE: 1.71% -> 0.45% | Helped |
| Add Mixup to AugMix BatchEnsemble | Wide ResNet-28-10 BatchEnsemble · CIFAR-100 | ECE: 2.96% -> 4.19% | Hurt |
| Replace uniform Mixup with class-adjusted Mixup in AugMix (ablation) | BatchEnsemble · CIFAR-100 | ECE: 4.19% -> 2.35% | Helped |
Closest papers by meaning
- Uncertainty Quantification and Deep Ensembles (NeurIPS 2021 · similarity 0.90)
- Better Uncertainty Calibration via Proper Scores for Classification and Beyond (NeurIPS 2022 · similarity 0.87)
- Beyond One-Hot Labels: Semantic Mixing for Model Calibration (ICML 2025 · similarity 0.85)
- Reassessing How to Compare and Improve the Calibration of Machine Learning Models (ICLR 2025 · similarity 0.85)
- Divide et Calibra: Multiclass Local Calibration via Vector Quantization (NeurIPS 2026 · similarity 0.84)
- Calibration Bottleneck: Over-compressed Representations are Less Calibratable (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.