Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
On Uncertainty, Tempering, and Data Augmentation in Bayesian Classification
Sanyam Kapoor, Wesley Maddox, Pavel Izmailov, Andrew Gordon Wilson
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 |
|---|---|---|---|
| Use a cold tempered softmax likelihood instead of the standard softmax likelihood | ResNet-18 Bayesian neural network · CIFAR-10 | test accuracy | Improved |
| Use the Noisy Dirichlet model instead of the standard softmax likelihood | ResNet-18 Bayesian neural network · CIFAR-10 | test accuracy | Improved |
| Model label noise with likelihood tempering or the Noisy Dirichlet model | ResNet-18 Bayesian neural network · CIFAR-10 and Tiny ImageNet with varying label noise | test accuracy | Improved |
| Use the Noisy Dirichlet model rather than tempered softmax | ResNet-18 Bayesian neural network · Tiny ImageNet with high label noise | test accuracy | Improved |
| Add data augmentation without compensating for its softened likelihood (ablation) | Bayesian neural network · CIFAR-10 train data | negative log-likelihood | Hurt |
| Add likelihood tempering to counteract data augmentation (ablation) | Bayesian neural network · Synthetic two-spirals classification | training fit | Improved |
| Use the Noisy Dirichlet model instead of tempering (ablation) | Bayesian neural network · Synthetic two-spirals classification | training fit | Improved |
| Use the Noisy Dirichlet model with posterior temperature 1 rather than adding posterior tempering (ablation) | ResNet-18 Bayesian neural network · CIFAR-10 | test accuracy | No clear effect |
| Use more complex data augmentation and lower the posterior temperature (ablation) | ResNet-18 Bayesian neural network · CIFAR-10 | test negative log-likelihood | Mixed |
| Use a proper augmentation likelihood rather than the standard likelihood (ablation) | Bayesian neural network · Image classification | performance | No clear effect |
| Use a smoothed softmax likelihood rather than a tempered likelihood (ablation) | ResNet-18 Bayesian neural network · CIFAR-10 | test performance | Hurt |
| Vary the prior at temperature 1 rather than tempering (ablation) | Bayesian neural network · CIFAR-10 | test accuracy | Hurt |
Closest papers by meaning
- Rethinking Aleatoric and Epistemic Uncertainty (ICML 2025 · similarity 0.90)
- How Tempering Fixes Data Augmentation in Bayesian Neural Networks (ICML 2022 · similarity 0.89)
- One Step Closer to Unbiased Aleatoric Uncertainty Estimation (AAAI 2024 · similarity 0.88)
- Estimating Epistemic and Aleatoric Uncertainty with a Single Model (NeurIPS 2024 · similarity 0.87)
- Is Epistemic Uncertainty Faithfully Represented by Evidential Deep Learning Methods? (ICML 2024 · similarity 0.87)
- Quantifying Aleatoric Uncertainty of the Treatment Effect: A Novel Orthogonal Learner (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.
