Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
On the Practicality of Deterministic Epistemic Uncertainty
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 SNGP instead of softmax entropy | Image classification model · CIFAR10-C and CIFAR100-C | AUROC: 0.782 -> 0.833; 0.762 -> 0.788 | Helped |
| Estimate uncertainty from hidden-representation density with DDU | Image classification model · CIFAR10-C and CIFAR100-C | rAULC: 0.441 vs 0.708; 0.339 vs 0.596 | Hurt |
| Use MC Dropout rather than the evaluated DUMs | Image classification model · CIFAR10-C and CIFAR100-C | rAULC: 0.829; 0.726 | Helped |
| Use SNGP instead of softmax entropy | Semantic segmentation model · Cityscapes-C and CARLA-C | rAULC: 0.737 -> 0.759; 0.818 -> 0.813 | Mixed |
| Estimate uncertainty from hidden-representation density with DDU | Semantic segmentation model · CARLA-C | rAULC: 0.818 vs -0.038 | Hurt |
| Increase MIR regularization strength (ablation) | MIR · MNIST rotated continuously from 0 to 180 degrees | rAULC correlation with regularization strength: Pearson 0.76; Spearman 0.96 | Helped |
| Increase DUQ regularization strength (ablation) | DUQ · MNIST rotated continuously from 0 to 180 degrees | rAULC correlation with regularization strength: Pearson -0.83; Spearman -0.83 | Mixed |
| Increase DDU regularization strength (ablation) | DDU · MNIST rotated continuously from 0 to 180 degrees | rAULC correlation with regularization strength: Pearson -0.30; Spearman -0.43 | No clear effect |
| Increase SNGP regularization strength (ablation) | SNGP · MNIST rotated continuously from 0 to 180 degrees | rAULC correlation with regularization strength: Pearson 0.58; Spearman 0.61 | Mixed |
| Scale DUMs to semantic segmentation | DUMs · Cityscapes-C and CARLA-C | mIoU | No clear effect |
| Apply DUQ to semantic segmentation | DUQ · Semantic segmentation | convergence | Hurt |
| Train DCU with contrastive pretraining | DCU · CIFAR10-C and CIFAR100-C | test accuracy: 0.945; 0.642 | Helped |
Closest papers by meaning
- Epistemic Uncertainty Quantification For Pre-Trained Neural Networks (CVPR 2024 · similarity 0.89)
- Is Epistemic Uncertainty Faithfully Represented by Evidential Deep Learning Methods? (ICML 2024 · similarity 0.88)
- Are Uncertainty Quantification Capabilities of Evidential Deep Learning a Mirage? (NeurIPS 2024 · similarity 0.87)
- Epistemic Uncertainty for Generated Image Detection (NeurIPS 2025 · similarity 0.87)
- CertainlyUncertain: A Benchmark and Metric for Multimodal Epistemic and Aleatoric Awareness (ICLR 2025 · similarity 0.87)
- Evaluating Epistemic Uncertainty: Beyond OOD Detection and Active Learning (NeurIPS 2026 · 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.