Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Deterministic Uncertainty: A New Simple Baseline
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 spectrally normalized Wide-ResNet-28-10 features with post-training GMM density | DDU · CIFAR-10 vs SVHN | AUROC: 94.44\pm0.43 -> 97.86\pm0.19 | Helped |
| Use spectrally normalized Wide-ResNet-28-10 features with post-training GMM density | DDU · CIFAR-10 vs CIFAR-100 | AUROC: 89.39\pm0.06 -> 91.34\pm0.04 | Helped |
| Use spectrally normalized Wide-ResNet-28-10 features with post-training GMM density | DDU · CIFAR-10 vs Tiny-ImageNet | AUROC: 88.42\pm0.05 -> 91.07\pm0.05 | Helped |
| Use DDU instead of the strongest listed ensemble result | Wide-ResNet-28-10 · CIFAR-10 vs CIFAR-100 | AUROC: 92.13\pm0.02 -> 91.34\pm0.04 | Hurt |
| Use DDU instead of the strongest listed ensemble result | Wide-ResNet-28-10 · CIFAR-10 vs SVHN | AUROC: 97.73\pm0.31 -> 97.86\pm0.19 | Helped |
| Add spectral normalization (ablation) | Wide-ResNet-28-10 DDU · CIFAR-10 vs SVHN | AUROC: 96.08\pm0.25 -> 97.86\pm0.19 | Helped |
| Add spectral normalization (ablation) | Wide-ResNet-28-10 DDU · CIFAR-10 vs CIFAR-100 | AUROC: 90.94\pm0.03 -> 91.34\pm0.04 | Helped |
| Replace Wide-ResNet-28-10 with VGG-16 (ablation) | DDU · CIFAR-10 vs SVHN | AUROC: 97.86\pm0.19 -> 89.62\pm0.37 | Hurt |
| Replace GMM feature density with softmax entropy for epistemic uncertainty (ablation) | Wide-ResNet-28-10 with spectral normalization · CIFAR-10 vs SVHN | AUROC: 97.86\pm0.19 -> 94.05\pm0.26 | Hurt |
| Add the GMM feature-density estimator (ablation) | Wide-ResNet-28-10 with spectral normalization · CIFAR-10 vs SVHN | AUROC: 94.31\pm0.33 -> 97.86\pm0.19 | Helped |
Closest papers by meaning
- On the Practicality of Deterministic Epistemic Uncertainty (ICML 2022 · similarity 0.92)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.88)
- Benchmarking Uncertainty Disentanglement: Specialized Uncertainties for Specialized Tasks (NeurIPS 2024 · similarity 0.88)
- Density-Softmax: Efficient Test-time Model for Uncertainty Estimation and Robustness under Distribution Shifts (ICML 2024 · similarity 0.88)
- Make Me a BNN: A Simple Strategy for Estimating Bayesian Uncertainty from Pre-trained Models (CVPR 2024 · similarity 0.87)
- One Step Closer to Unbiased Aleatoric Uncertainty Estimation (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.