Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
On the Pitfalls of Heteroscedastic Uncertainty Estimation with Probabilistic Neural Networks
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 |
|---|---|---|---|
| Weight per-sample NLL contributions by variance to the power β, using β=0.5 rather than standard NLL (ablation) | Probabilistic dynamics models · 1D-Slide | RMSE: 0.0192 ± 0.006 -> 0.0064 ± 0.002 | Helped |
| Use β-NLL with β=0.5 instead of Student-t | Probabilistic dynamics models · 1D-Slide | LL: 11.30 ± 0.03 vs 9.28 ± 0.75 | Hurt |
| Use β-NLL with β=0.5 instead of standard NLL (ablation) | Probabilistic dynamics models · Fetch-PickAndPlace | RMSE: 0.00163 ± 0.00008 -> 0.00096 ± 0.00002 | Helped |
| Use β-NLL with β=0.5 instead of Student-t | Probabilistic dynamics models · Fetch-PickAndPlace | LL: 30.44 ± 0.08 vs 24.68 ± 0.08 | Hurt |
| Set β to 0.5 instead of 0 in β-NLL (ablation) | VAE decoder · MNIST | RMSE: 0.237 ± 0.002 -> 0.151 ± 0.003 | Helped |
| Set β to 0.5 instead of 0 in β-NLL (ablation) | VAE decoder · Fashion-MNIST | RMSE: 0.170 ± 0.001 -> 0.125 ± 0.003 | Helped |
| Use β-NLL with β=0.5 instead of Student-t | VAE decoder · MNIST | LL: 4291 ± 103 vs 2220 ± 25 | Hurt |
| Use β-NLL with β=0.5 instead of standard NLL (ablation) | AdaBins without Mini-ViT Transformer · NYUv2 | RMSE: 0.3854 -> 0.3789 | Helped |
| Use β-NLL with β=0.5 instead of MSE | AdaBins without Mini-ViT Transformer · NYUv2 | RMSE: 0.3776 vs 0.3789 | Hurt |
| Set β to 0.75 instead of β=0.5 (ablation) | β-NLL · 1D-Slide | RMSE: 0.0064 ± 0.002 -> 0.0087 ± 0.003 | Hurt |
Closest papers by meaning
- Effective Bayesian Heteroscedastic Regression with Deep Neural Networks (NeurIPS 2023 · similarity 0.90)
- Heteroscedastic Variational Last Layers (NeurIPS 2026 · similarity 0.88)
- Deep Heteroskedastic Regression: Post-Hoc Variance Estimation from Latent Representations (NeurIPS 2026 · similarity 0.88)
- Uncertainty Quantification with the Empirical Neural Tangent Kernel (NeurIPS 2025 · similarity 0.87)
- One Step Closer to Unbiased Aleatoric Uncertainty Estimation (AAAI 2024 · similarity 0.86)
- Estimating Epistemic and Aleatoric Uncertainty with a Single Model (NeurIPS 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.