Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Revisiting Essential and Nonessential Settings of Evidential Deep Learning
Mengyuan Chen, Junyu Gao, Changsheng Xu
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 |
|---|---|---|---|
| replace softmax head with projected probability (MSE loss) (ablation) | Re-EDL · CIFAR-10 (VGG16) mean OOD detection | AUPR: 81.30 -> 85.46 | Improved |
| replace softmax head with projected probability (CE loss) (ablation) | Re-EDL · CIFAR-10 (VGG16) mean OOD detection | AUPR: 80.33 -> 84.56 | Improved |
| use softplus evidence instead of ReLU (ablation) | Re-EDL · CIFAR-10 (VGG16) mean OOD detection | AUPR: 81.81 -> 85.46 | Improved |
| use softplus evidence instead of Exp (ablation) | Re-EDL · CIFAR-10 (VGG16) mean OOD detection | AUPR: 82.39 -> 85.46 | Improved |
| add KL regularization on non-target evidence (ablation) | EDL · CIFAR-10 (VGG16) mean OOD detection | AUPR: 84.50 -> 82.32 | Hurt |
| add KL regularization on non-target evidence (ablation) | Re-EDL · CIFAR-10 (VGG16) mean OOD detection | AUPR: 85.46 -> 83.73 | Hurt |
| add KL regularization (R-EDL vs Re-EDL) under Gaussian noise SD 0.200 (ablation) | Re-EDL · CIFAR-10 noisy test | average score: 54.83 -> 57.16 | Improved |
| add KL regularization (R-EDL vs Re-EDL) in few-shot 5-way 5-shot (ablation) | Re-EDL · mini-ImageNet to CUB OOD-UM | AUPR: 83.35 -> 84.22 | Improved |
| add small KL coefficient (lg mu -3.0) (ablation) | Re-EDL · CIFAR-10 (VGG16) mean OOD detection | AUPR: 85.46 -> 84.41 | Hurt |
| add Re-EDL to deep ensemble of 5 | Deep Ensemble(5) · CIFAR-10 (VGG16) mean OOD detection | AUPR: 84.47 -> 88.35 | Improved |
| add Re-EDL to deep ensemble of 25 | Deep Ensemble(25) · CIFAR-10 (VGG16) mean OOD detection | AUPR: 85.88 -> 88.96 | Improved |
| replace EDL with R-EDL (relax prior weight, drop variance term) | EDL · CIFAR-10 (VGG16) mean OOD detection | AUPR: 82.32 -> 83.73 | Improved |
Closest papers by meaning
- Uncertainty Estimation by Flexible Evidential Deep Learning (NeurIPS 2025 · similarity 0.93)
- Are Uncertainty Quantification Capabilities of Evidential Deep Learning a Mirage? (NeurIPS 2024 · similarity 0.93)
- Calibrating LLMs with Information-Theoretic Evidential Deep Learning (ICLR 2025 · similarity 0.90)
- Uncertainty Estimation by Density Aware Evidential Deep Learning (ICML 2024 · similarity 0.90)
- Is Epistemic Uncertainty Faithfully Represented by Evidential Deep Learning Methods? (ICML 2024 · similarity 0.89)
- Uncertainty Regularized Evidential Regression (AAAI 2024 · similarity 0.87)
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.
