Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Estimating Regression Predictive Distributions with Sample Networks
Ali Harakeh, Jordan Sir Kwang Hu, Naiqing Guan, Steven Waslander, Liam Paull
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 |
|---|---|---|---|
| represent predictions as learned samples and train with the Energy Score instead of a parametric distribution | SampleNet · Weather | Energy Score: 0.594 ± 0.032 -> 0.106 ± 0.002 | Improved |
| represent predictions as learned samples and train with the Energy Score instead of a parametric distribution | SampleNet · Traffic | Energy Score: 133.884 ± 3.138 -> 12.562 ± 0.389 | Improved |
| use learned samples with flexible predictive distributions | SampleNet · NYUv2 | Gaussian NLL: 7.237 ± 6.85e-1 -> 1.398 ± 1.22e-3 | Improved |
| use Sinkhorn regularization with η=1.0 (ablation) | SampleNet · NYUv2 | Average variance, All: 0.0157 -> 0.0115 | Improved |
| remove Sinkhorn regularization (η=0) (ablation) | SampleNet · NYUv2 | Energy Score: 0.2371 ± 5.43e-4 -> 0.2354 ± 4.01e-3 | Hurt |
| increase Sinkhorn regularization strength from η=0 to η=5 (ablation) | SampleNet · Weather | Energy Score: 0.117 ± 0.002 -> 0.106 ± 0.002 | Improved |
| increase Sinkhorn regularization strength (ablation) | SampleNet · Traffic | Energy Score | No clear effect |
| increase minibatch loss sample size K (ablation) | SampleNet · Traffic | Energy Score | Improved |
| increase the number of minibatch loss repetitions L (ablation) | SampleNet · Traffic | Energy Score | Improved |
| increase the number of predicted samples M (ablation) | SampleNet · Weather and Traffic | Energy Score | No clear effect |
| increase output samples from M=25 to M=50 (ablation) | SampleNet · NYUv2 | Energy Score: 0.2371 -> 0.2381 | Hurt |
| remove Sinkhorn regularization (ablation) | SampleNet · NYUv2 | δ1: 0.8900 -> 0.8891 | Hurt |
Closest papers by meaning
- Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles (NeurIPS 2017 · similarity 0.90)
- Distribution-Free Data Uncertainty for Neural Network Regression (ICLR 2025 · similarity 0.89)
- Uncertainty Quantification for Deep Regression using Contextualised Normalizing Flows (NeurIPS 2025 · similarity 0.87)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.86)
- Uncertainty Quantification with the Empirical Neural Tangent Kernel (NeurIPS 2025 · similarity 0.86)
- Quantifying Uncertainty in the Presence of Distribution Shifts (NeurIPS 2025 · 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.
