Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Improving model calibration with accuracy versus uncertainty optimization
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 AvUC loss to SVI training (SVI-AvUC vs SVI) (ablation) | SVI ResNet-50 · ImageNet shift intensity 5 | mean ECE: 0.1188 -> 0.0542 | Helped |
| add AvUC loss to SVI training (ablation) | SVI ResNet-50 · ImageNet shift intensity 5 | mean UCE: 0.2865 -> 0.1774 | Helped |
| post-hoc AvUTS temperature scaling on SVI (ablation) | SVI ResNet-50 · ImageNet shift intensity 5 | mean ECE: 0.1188 -> 0.0800 | Helped |
| add AvUC loss to SVI training (ablation) | SVI ResNet-50 · ImageNet shift intensity 5 | mean NLL: 4.865 -> 4.707 | Helped |
| add AvUC loss to SVI (ablation) | SVI · ImageNet dataset shift detection | AUROC: 96.40 -> 97.60 | Helped |
| add AvUC loss to SVI (ablation) | SVI · CIFAR10 dataset shift detection | AUROC: 85.89 -> 95.54 | Helped |
| add AvUC loss to SVI (ablation) | SVI · CIFAR10 OOD detection (SVHN) | AUROC: 93.94 -> 99.35 | Helped |
| post-hoc AvUTS instead of NLL temperature scaling on SVI (ablation) | SVI · CIFAR10 dataset shift detection | AUROC: 81.08 -> 81.19 | No clear effect |
| add AvUC loss to vanilla DNN (ablation) | Vanilla ResNet-20 · CIFAR10 shift intensity 5 | ECE (%): 31.5 -> 23.2 | Helped |
| add AvUC loss to vanilla DNN (ablation) | Vanilla ResNet-20 · CIFAR10 shift intensity 5 | UCE (%): 28.5 -> 16.8 | Helped |
| AvUTS post-hoc on vanilla DNN (ablation) | Vanilla ResNet-20 · CIFAR10 in-distribution (intensity 0) | ECE (%): 4.6 -> 2.1 | Helped |
| AvUC on vanilla vs vanilla in-distribution (ablation) | Vanilla ResNet-20 · CIFAR10 in-distribution (intensity 0) | ECE (%): 4.6 -> 3.4 | Helped |
Closest papers by meaning
- Uncertainty Weighted Gradients for Model Calibration (CVPR 2025 · similarity 0.92)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.89)
- A Benchmark Study on Calibration (ICLR 2024 · similarity 0.89)
- Provable Uncertainty Decomposition via Higher-Order Calibration (ICLR 2025 · similarity 0.88)
- Calibration Bottleneck: Over-compressed Representations are Less Calibratable (ICML 2024 · similarity 0.88)
- Uncertainty Quantification with the Empirical Neural Tangent Kernel (NeurIPS 2025 · similarity 0.88)
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.