Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
How Tempering Fixes Data Augmentation in Bayesian Neural Networks
Gregor Bachmann, Lorenzo Noci, Thomas Hofmann
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 |
|---|---|---|---|
| Temper only the likelihood instead of the full posterior | ResNet20 · CIFAR-10 with random flips and crops | test cross-entropy | No clear effect |
| Replace standard convolutions with group convolutions | ResNet20 · CIFAR-10 with crops and flips, with or without 90° rotations | test cross-entropy | Improved |
| Add 90° rotations to the augmentation set (ablation) | ResNet20 and G-ResNet · CIFAR-10 | test cross-entropy | Hurt |
| Add 90° rotations to the augmentation set (ablation) | ResNet20 · CIFAR-10 | optimal temperature | Mixed |
| Add 90° rotations to the augmentation set (ablation) | G-ResNet · CIFAR-10 | optimal temperature | No clear effect |
| Use stride 1 rather than stride 2 (ablation) | G-convolutional network · CIFAR-10 with flips and 90° rotations | test cross-entropy at T=150 | No clear effect |
| Add random 10° rotations (ablation) | G-convolutional network · CIFAR-10 | optimal temperature | Hurt |
Closest papers by meaning
- Disentangling the Roles of Curation, Data-Augmentation and the Prior in the Cold Posterior Effect (NeurIPS 2021 · similarity 0.90)
- On Uncertainty, Tempering, and Data Augmentation in Bayesian Classification (NeurIPS 2022 · similarity 0.89)
- Quantifying Uncertainty in the Presence of Distribution Shifts (NeurIPS 2025 · similarity 0.85)
- A Study of Bayesian Neural Network Surrogates for Bayesian Optimization (ICLR 2024 · similarity 0.84)
- A Symmetry-Aware Exploration of Bayesian Neural Network Posteriors (ICLR 2024 · similarity 0.84)
- Effortless, Simulation-Efficient Bayesian Inference using Tabular Foundation Models (NeurIPS 2025 · similarity 0.84)
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.
