Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Riemannian Laplace approximations for Bayesian neural 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 |
|---|---|---|---|
| Use a loss-aware Riemannian metric and exponential-map sampling instead of Euclidean Laplace sampling | Laplace approximation · UCI classification datasets | NLL: Vehicle: 1.209 ± 0.020 -> 0.454 ± 0.024; Glass: 1.737 ± 0.037 -> 1.047 ± 0.224; Ionosphere: 0.673 ± 0.013 -> 0.344 ± 0.0 | Helped |
| Use a loss-aware Riemannian metric and exponential-map sampling | Laplace approximation · MNIST | NLL: 0.871 ± 0.026 -> 0.115 ± 0.003 | Helped |
| Use a loss-aware Riemannian metric and exponential-map sampling | Laplace approximation · FashionMNIST | Accuracy: 79.88 ± 0.09 -> 83.33 ± 0.17 | Helped |
| Use a loss-aware Riemannian metric and exponential-map sampling | Laplace approximation · banana dataset, prior optimized | NLL: 0.678 ± 0.009 -> 0.287 ± 0.002 | Helped |
| Leave the prior precision unoptimized (ablation) | Riemannian LA · banana dataset | NLL: 0.287 ± 0.002 -> 0.285 ± 0.001 | No clear effect |
| Use mini-batches when solving the exponential-map ODE (ablation) | Riemannian LA · MNIST | NLL: 0.115 ± 0.003 -> 0.170 ± 0.005 | Hurt |
| Linearize the model used to construct the Riemannian geometry (ablation) | Riemannian LA · banana dataset, prior optimized | NLL: 0.287 ± 0.002 -> 0.415 ± 0.029 | Hurt |
| Add mini-batches to linearized Riemannian LA (ablation) | Linearized Riemannian LA · banana dataset, prior optimized | NLL: 0.415 ± 0.029 -> 0.298 ± 0.006 | Helped |
| Use the linearized model without batches (ablation) | Riemannian LA · FashionMNIST | NLL: 0.472 ± 0.001 -> 0.521 ± 0.004 | Hurt |
| Optimize prior precision for vanilla LA (ablation) | Vanilla LA · banana dataset | Accuracy: 48.85 ± 2.32 -> 59.50 ± 5.07 | Helped |
| Use the linearized model for Laplace sampling (ablation) | Vanilla Laplace approximation · banana dataset, prior optimized | Accuracy: 59.50 ± 5.07 -> 86.99 ± 0.37 | Helped |
Closest papers by meaning
- Laplace Redux - Effortless Bayesian Deep Learning (NeurIPS 2021 · similarity 0.90)
- FSP-Laplace: Function-Space Priors for the Laplace Approximation in Bayesian Deep Learning (NeurIPS 2024 · similarity 0.89)
- Reparameterization invariance in approximate Bayesian inference (NeurIPS 2024 · similarity 0.88)
- Riemannian Stein Variational Gradient Descent for Bayesian Inference (AAAI 2018 · similarity 0.88)
- RMLR: Extending Multinomial Logistic Regression into General Geometries (NeurIPS 2024 · similarity 0.87)
- Riemannian Multinomial Logistics Regression for SPD Neural Networks (CVPR 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.