Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Sampling-based inference for large linear models, with application to linearised Laplace
Javier Antoran, Shreyas Padhy, Riccardo Barbano, Eric Nalisnick, David Janz, José Miguel Hernández-Lobato
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 the sample-based linearised Laplace method instead of lin.-UNet | U-Net · Tomographic reconstruction, m=7680 | LL (first column): 2.214 -> 2.341 | Improved |
| Use the sample-based linearised Laplace method instead of lin.-UNet | U-Net · Tomographic reconstruction, m=7680 | LL (second column): 2.601 -> 2.869 | Improved |
| Use the sample-based linearised Laplace method instead of MCDO-UNet | U-Net · Tomographic reconstruction, m=7680 | LL (first column): 0.028 -> 2.341 | Improved |
| Use the sample-based linearised Laplace method instead of MCDO-UNet | U-Net · Tomographic reconstruction, m=7680 | LL (second column): 2.474 -> 2.869 | Improved |
| Use the sample-based linearised Laplace method instead of MCDO-UNet | U-Net · Tomographic reconstruction, m=15360 | LL (first column): 0.002 -> 2.310 | Improved |
| Use the sample-based linearised Laplace method instead of MCDO-UNet | U-Net · Tomographic reconstruction, m=15360 | LL (second column): 2.762 -> 2.972 | Improved |
| Replace the standard sampling objective L with the proposed objective L' (ablation) | Sample-based linearised Laplace · MNIST test set, LeNet | sample fidelity | Improved |
| Replace the standard sampling objective L with the proposed objective L' (ablation) | Sample-based linearised Laplace · MNIST test set, LeNet | predictive error | Improved |
| Use a single prior precision instead of layer-wise prior precisions (ablation) | Sampling EM · MNIST, LeNet | EM convergence: about 3 steps | Improved |
| Use layer-wise prior precisions (ablation) | Sampling EM · MNIST, LeNet | EM convergence: within 15 EM steps | Hurt |
| Use the g-prior with a single regularisation parameter instead of separate parameters for normalised layers (ablation) | Linearised Laplace with normalisation layers · Neural-network Jacobian features | model evidence objective well-definedness | Improved |
Closest papers by meaning
- A Study of Bayesian Neural Network Surrogates for Bayesian Optimization (ICLR 2024 · similarity 0.86)
- FSP-Laplace: Function-Space Priors for the Laplace Approximation in Bayesian Deep Learning (NeurIPS 2024 · similarity 0.86)
- Reparameterization invariance in approximate Bayesian inference (NeurIPS 2024 · similarity 0.86)
- Microcanonical Langevin Ensembles: Advancing the Sampling of Bayesian Neural Networks (ICLR 2025 · similarity 0.86)
- Provable Posterior Sampling with Denoising Oracles via Tilted Transport (NeurIPS 2024 · similarity 0.86)
- Effortless, Simulation-Efficient Bayesian Inference using Tabular Foundation Models (NeurIPS 2025 · 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.
