Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Liberty or Depth: Deep Bayesian Neural Nets Do Not Need Complex Weight Posterior Approximations
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 |
|---|---|---|---|
| Use diagonal covariance instead of MVG covariance | ResNet-18 · ImageNet | accuracy: 66.4% -> 67.4% | Helped |
| Use diagonal covariance instead of MVG covariance | ResNet-18 · ImageNet | NLL: 1.44 -> 1.37 | Helped |
| Use diagonal covariance instead of MVG covariance | ResNet-18 · ImageNet | ECE: 0.080 -> 0.029 | Helped |
| Use diagonal covariance instead of low-rank structured covariance | DenseNet-161 SWAG · ImageNet | accuracy: 78.6% -> 78.6% | No clear effect |
| Use diagonal covariance instead of low-rank structured covariance | DenseNet-161 SWAG · ImageNet | NLL: 0.83 -> 0.86 | Hurt |
| Use diagonal covariance instead of low-rank structured covariance | DenseNet-161 SWAG · ImageNet | ECE: 0.020 -> 0.046 | Hurt |
| Use diagonal covariance instead of low-rank structured covariance | ResNet-152 SWAG · ImageNet | accuracy: 79.1% -> 80.0% | Helped |
| Use diagonal covariance instead of low-rank structured covariance | ResNet-152 SWAG · ImageNet | NLL: 0.82 -> 0.86 | Hurt |
| Use diagonal covariance instead of low-rank structured covariance | ResNet-152 SWAG · ImageNet | ECE: 0.028 -> 0.057 | Hurt |
| Add depth to mean-field networks | Mean-field BNN · Iris | cross-entropy | Mixed |
| Add depth to mean-field networks | Mean-field BNN · two-moons | Wasserstein error | Helped |
| Add depth to mean-field networks | Mean-field BNN · two-moons | KL-divergence error | Helped |
Closest papers by meaning
- On the Expressiveness of Approximate Inference in Bayesian Neural Networks (NeurIPS 2020 · similarity 0.92)
- The k-tied Normal Distribution: A Compact Parameterization of Gaussian Mean Field Posteriors in Bayesian Neural Networks (ICML 2020 · similarity 0.91)
- VIKING: Deep variational inference with stochastic projections (NeurIPS 2025 · similarity 0.91)
- Structured and Efficient Variational Deep Learning with Matrix Gaussian Posteriors (ICML 2016 · similarity 0.90)
- Global inducing point variational posteriors for Bayesian neural networks and deep Gaussian processes (ICML 2021 · similarity 0.90)
- Variational Bayesian Pseudo-Coreset (ICLR 2025 · 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.