Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Bayesian Uncertainty Estimation for Batch Normalized Deep 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 |
|---|---|---|---|
| Estimate predictive uncertainty using stochastic BN mini-batches instead of fixed population statistics | Batch-normalized networks · CIFAR10 test | PLL: -0.32 -> -0.28 | Helped |
| Increase stochastic forward passes from 1 to 128 (ablation) | MCBN / ResNet32 · CIFAR10 test | PLL: -.36 -> -.28 | Helped |
| Increase stochastic forward passes beyond 32 (ablation) | MCBN / ResNet32 · CIFAR10 test | PLL: -.28 -> -.28 | No clear effect |
| Use very small BN batch sizes (≤16) (ablation) | MCBN · Regression datasets | uncertainty quality | Hurt |
| Use fewer stochastic passes, such as T=50 (ablation) | MCBN · Regression datasets | uncertainty quality | Mixed |
| Use a larger inference mini-batch (ablation) | MCBN applied to Bayesian SegNet · PASCAL-VOC vs CamVid | qualitative uncertainty quality | Helped |
| Apply MCBN instead of constant uncertainty | MCBN · Eight regression datasets | normalized CRPS and normalized PLL | Helped |
| Use MCBN rather than MCDO or MNF | Approximate Bayesian uncertainty estimation · Eight regression datasets | normalized uncertainty quality | Mixed |
Closest papers by meaning
- Conservative Uncertainty Estimation By Fitting Prior Networks (ICLR 2020 · similarity 0.90)
- Modeling Uncertainty by Learning a Hierarchy of Deep Neural Connections (NeurIPS 2019 · similarity 0.90)
- Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles (NeurIPS 2017 · similarity 0.90)
- Simple and Principled Uncertainty Estimation with Deterministic Deep Learning via Distance Awareness (NeurIPS 2020 · similarity 0.89)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.88)
- Make Me a BNN: A Simple Strategy for Estimating Bayesian Uncertainty from Pre-trained Models (CVPR 2024 · similarity 0.88)
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.