Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
SLANG: Fast Structured Covariance Approximations for Bayesian Deep Learning with Natural Gradient
Aaron Mishkin, Frederik Kunstner, Didrik Nielsen, Mark Schmidt, Mohammad Emtiyaz Khan
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 low-rank plus diagonal covariance approximation instead of a mean-field covariance | SLANG Bayesian logistic regression · Australian, Breast Cancer, a1a, USPS 3vs5 | ELBO: 0.614 -> 0.566; 0.122 -> 0.111; 0.384 -> 0.373; 0.268 -> 0.193 | Improved |
| Use SLANG instead of BBB | Bayesian neural networks · UCI regression datasets | Test RMSE: outperforms BBB on 7 out of 8 datasets | Improved |
| Use SLANG instead of Bayesian Dropout | Bayesian neural networks · UCI regression datasets | Test RMSE | Mixed |
| Increase the SLANG covariance rank (ablation) | Bayesian logistic regression · Australian, Breast Cancer, a1a, USPS 3vs5 | ELBO: Australian: 0.574 -> 0.569; Breast Cancer: 0.112 -> 0.111; a1a: 0.377 -> 0.374; USPS 3vs5: 0.210 -> 0.198 | Improved |
| Increase the SLANG covariance rank (ablation) | SLANG Bayesian neural network · MNIST | Test Error: 2.00% -> 1.73%; L=8: 1.92% | Mixed |
| Replace the SLANG precision-update projection with online eigendecomposition and moving-average factor updates (ablation) | SLANG-OnlineEig · Australian, Breast Cancer, a1a, USPS 3vs5 | ELBO: Australian L=5: 0.569 -> 0.652; Breast Cancer L=5: 0.111 -> 0.123; a1a L=5: 0.374 -> 0.383 | Hurt |
| Increase SLANG-OnlineEig rank from L=2 to L=5 (ablation) | SLANG-OnlineEig · Australian | ELBO: 0.578 -> 0.652 | Hurt |
| Use SLANG rather than the mean-field method | Bayesian neural networks · USPS 3vs5 and Breast Cancer convergence experiments | ELBO convergence speed | Improved |
Closest papers by meaning
- Noisy Natural Gradient as Variational Inference (ICML 2018 · similarity 0.88)
- Bayesian Online Natural Gradient (BONG) (NeurIPS 2024 · similarity 0.86)
- VIKING: Deep variational inference with stochastic projections (NeurIPS 2025 · similarity 0.86)
- Sparse Inducing Points in Deep Gaussian Processes: Enhancing Modeling with Denoising Diffusion Variational Inference (ICML 2024 · similarity 0.85)
- Understanding Stochastic Natural Gradient Variational Inference (ICML 2024 · similarity 0.85)
- Variational Linearized Laplace Approximation for Bayesian Deep Learning (ICML 2024 · similarity 0.85)
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.
