Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Evidential Transformation Network: Turning Pretrained Models into Evidential Models for Post-hoc Uncertainty Estimation
Yongchan Chun, Chanhee Park, Jeongho Yoon, Jaehyung Seo, Heuiseok Lim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 ETN instead of fine-tuning with CE loss on ImageNet | ResNet50 (ImageNet) · ImageNet | accuracy: 46.03 -> 79.61 | Improved |
| use ETN instead of MAP_EDL fine-tuning on CIFAR-10 OOD (SVHN) | VGG16 · CIFAR-10 -> SVHN | AUPR (MI): 76.42 -> 85.22 | Improved |
| use ETN instead of Laplace approximation on ImageNet-A OOD | ResNet50 · ImageNet -> ImageNet-A | AUPR (MI): 96.36 -> 93.98 | Hurt |
| use sample-dependent ETN transformation instead of static scaling or AdaTS on ImageNet | ETN · ImageNet | Uncertainty Mass AUPR: about 17% and 10% drops in UM for static scaling and AdaTS | Improved |
| use ETN instead of static scaling or AdaTS for OOD mutual information on ImageNet | ETN · ImageNet | AUPR (MI) OOD: 21.5% and 25.5% lower MI for static scaling and AdaTS | Improved |
| use ETN instead of static scaling or AdaTS for mutual information on OBQA (ablation) | ETN · OBQA | AUPR (MI) OOD: 13.6% and 24.7% lower MI for static scaling and AdaTS | Improved |
| use ETN instead of static scaling on CIFAR-10 mutual information (ablation) | ETN · CIFAR-10 | AUPR (MI) OOD: static scaling trails ETN by roughly 5% | Improved |
| make the transformation sample-dependent rather than static (ablation) | ETN / AdaTS vs static scaling · ImageNet | uncertainty estimation | Improved |
| use scalar instead of vector or matrix transformation parameter (ablation) | ETN · OBQA | ID and OOD AUPR | Improved |
| use matrix instead of scalar transformation parameter (ablation) | ETN · CIFAR-10 | OOD AUPR | Mixed |
| use vector instead of scalar parameter (ablation) | ETN · RACE vs ImageNet | AUPR | Mixed |
| increase the mode of the prior p(A) (ablation) | ETN · CIFAR-10 and RACE | AUPR | Improved |
Closest papers by meaning
- Uncertainty Regularized Evidential Regression (AAAI 2024 · similarity 0.89)
- Uncertainty Estimation by Flexible Evidential Deep Learning (NeurIPS 2025 · similarity 0.89)
- Are Uncertainty Quantification Capabilities of Evidential Deep Learning a Mirage? (NeurIPS 2024 · similarity 0.88)
- Calibrating LLMs with Information-Theoretic Evidential Deep Learning (ICLR 2025 · similarity 0.87)
- Hyper Evidential Deep Learning to Quantify Composite Classification Uncertainty (ICLR 2024 · similarity 0.87)
- Make Me a BNN: A Simple Strategy for Estimating Bayesian Uncertainty from Pre-trained Models (CVPR 2024 · 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.
