Research map / Label-efficient and robust learning / Recognition and long-tail
Post-hoc Uncertainty Learning Using a Dirichlet Meta-Model
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 a Dirichlet Bayesian meta-model with multiple intermediate-layer features for OOD uncertainty instead of base-model uncertainty scores | LeNet on MNIST · MNIST → Omniglot | AUROC: 98.9$\pm$0.5 -> 99.9$\pm$0.0 | Helped |
| Use epistemic uncertainty scores from the Dirichlet meta-model for OOD detection | VGG16 on CIFAR10 · CIFAR10 → SVHN | AUROC: 86.4$\pm$4.6 -> 100.0$\pm$0.0 | Helped |
| Use the proposed method for misclassification detection | LeNet on MNIST · MNIST misclassification | AUPR: 39.4$\pm$3.6 -> 44.5$\pm$5.1 | Helped |
| Use the proposed method for misclassification detection | VGG16 on CIFAR10 · CIFAR10 misclassification | AUROC: 94.5$\pm$0.0 -> 92.2$\pm$0.7 | Hurt |
| Adapt the proposed meta-model to a target transfer task | ImageNet-pretrained ResNet-50 · STL10 transfer, FashionMNIST OOD | AUROC: 89.2$\pm$0.9 -> 91.8$\pm$0.8 | Helped |
| Use multiple intermediate-layer features instead of only the final-layer feature (ablation) | Dirichlet meta-model on CIFAR10 · CIFAR10 → SVHN | AUROC: 90.6 -> 100.0 | Helped |
| Use the Dirichlet objective instead of standard cross-entropy for meta-model training (ablation) | Meta-model on CIFAR10 · CIFAR10 → SVHN | AUROC: 94.2 -> 100.0 | Helped |
| Retrain only the base model's last layer using cross-entropy and the proposed validation trick (ablation) | VGG16 on CIFAR10 · CIFAR10 → SVHN | AUROC: 86.4 -> 93.0 | Helped |
| Train the meta-model using only 10% of the training samples (ablation) | Dirichlet meta-model on CIFAR10 · CIFAR10 → SVHN | AUROC: 99.9 vs 100.0 | No clear effect |
Closest papers by meaning
- Towards Maximizing the Representation Gap between In-Domain & Out-of-Distribution Examples (NeurIPS 2020 · similarity 0.88)
- Dirichlet-based Uncertainty Calibration for Active Domain Adaptation (ICLR 2023 · similarity 0.87)
- Multidimensional Belief Quantification for Label-Efficient Meta-Learning (CVPR 2022 · similarity 0.87)
- Uncertainty-Informed Meta Pseudo Labeling for Surrogate Modeling with Limited Labeled Data (NeurIPS 2025 · similarity 0.87)
- Stop Guessing: Choosing the Optimization-Consistent Uncertainty Measurement for Evidential Deep Learning (ICLR 2026 · similarity 0.87)
- Uncertainty Estimation by Flexible Evidential Deep Learning (NeurIPS 2025 · similarity 0.86)
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.