Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Adapting the Linearised Laplace Model Evidence for Modern Deep Learning
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 the linear-model optimum instead of the neural-network linearization point for evidence optimization | Linearised Laplace · pointcloud-MNIST | test NLL: 0.310 -> 0.162 | Helped |
| Use the linear-model optimum instead of the neural-network linearization point for evidence optimization | Linearised Laplace · MNIST | test NLL: 0.253 -> 0.025 | Helped |
| Use the linear-model optimum instead of the neural-network linearization point for evidence optimization | Linearised Laplace · MNIST | test NLL: 0.252 -> 0.017 | Helped |
| Use the linear-model optimum instead of the neural-network linearization point for evidence optimization | Linearised Laplace · MNIST | test NLL: 0.220 -> 0.017 | Helped |
| Use the linear-model optimum instead of the neural-network linearization point for evidence optimization | Linearised Laplace · MNIST | test NLL: 0.153 -> 0.055 | Helped |
| Use the linear-model optimum instead of the neural-network linearization point for evidence optimization | Linearised Laplace · KMNIST tomographic reconstruction | test NLL: -1.164 -> -1.793 | Helped |
| Use layerwise rather than single regularization for FixUp ResNet (ablation) | Linearised Laplace with Mθ* · MNIST | test NLL: 0.055 -> 0.061 | Hurt |
| Use layerwise rather than single regularization (ablation) | Linearised Laplace with Mθ* · KMNIST tomographic reconstruction | test NLL: -1.793 -> -2.240 | Helped |
| Use simplified evidence with the linear-model optimum rather than evidence at the neural-network linearization point | ResNet-50 with batch norm · CIFAR10 | test NLL: 0.570 -> 0.112 | Helped |
| Use simplified evidence with the linear-model optimum rather than evidence at the neural-network linearization point | FixUp ResNet-50 · CIFAR10 | test NLL: 0.412 -> 0.128 | Helped |
| Use layerwise rather than single regularization (ablation) | ResNet-50 with batch norm using Mθ*,simple · CIFAR10 | test NLL: 0.112 -> 0.109 | Helped |
| Use layerwise rather than single regularization (ablation) | FixUp ResNet-50 using Mθ*,simple · CIFAR10 | test NLL: 0.128 -> 0.096 | Helped |
Closest papers by meaning
- FSP-Laplace: Function-Space Priors for the Laplace Approximation in Bayesian Deep Learning (NeurIPS 2024 · similarity 0.90)
- Variational Linearized Laplace Approximation for Bayesian Deep Learning (ICML 2024 · similarity 0.88)
- Debiasing Mini-Batch Quadratics for Applications in Deep Learning (ICLR 2025 · similarity 0.86)
- Bayesian Low-Rank Learning (Bella): A Practical Approach to Bayesian Neural Networks (AAAI 2025 · similarity 0.85)
- Reparameterization invariance in approximate Bayesian inference (NeurIPS 2024 · similarity 0.85)
- Laplace Sample Information: Data Informativeness Through a Bayesian Lens (ICLR 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.