Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Should We Learn Most Likely Functions or Parameters?
Shikai Qiu, Tim G. J. Rudner, Sanyam Kapoor, Andrew Gordon Wilson
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 |
|---|---|---|---|
| function-space MAP (Laplacian approximation) instead of parameter-space MAP | MLP 3x256 · UCI regression (8 datasets) | normalized test RMSE: .032 -> .018 | Improved |
| L-MAP instead of PS-MAP | MLP 3x256 · UCI Boston | normalized test RMSE: .329 -> .352 | Hurt |
| L-MAP instead of PS-MAP | ResNet-18 · CIFAR-10 | accuracy: 95.4 -> 95.3 | No clear effect |
| L-MAP with CIFAR-100 evaluation points instead of PS-MAP | ResNet-18 · CIFAR-10 | ECE: 2.5 -> 1.4 | Improved |
| L-MAP instead of PS-MAP in ImageNet transfer | ResNet-50 · CIFAR-10 | NLL: 0.18 -> 0.15 | Improved |
| L-MAP instead of PS-MAP in ImageNet transfer | ResNet-50 · CIFAR-10 | ECE: 2.6 -> 2.1 | Improved |
| L-MAP with KMNIST evaluation points instead of PS-MAP | ResNet-18 · FashionMNIST | accuracy: 93.9 -> 94.1 | No clear effect |
| use external evaluation set (CIFAR-100) rather than N(0,I) noise for evaluation distribution (ablation) | L-MAP · CIFAR-10 | NLL: .20 -> .16 | Improved |
| increase Monte Carlo samples S for Laplacian estimator (ablation) | L-MAP · CIFAR-10 | accuracy | No clear effect |
| increase Monte Carlo samples S for Laplacian estimator (ablation) | L-MAP · CIFAR-10 | calibration error | Improved |
| training with label noise: L-MAP vs PS-MAP | L-MAP · CIFAR-10 | test performance under label noise | Improved |
| increase number of evaluation points up to 200 (ablation) | FS-MAP · synthetic Fourier regression, 400 training samples | test RMSE | Improved |
Closest papers by meaning
- Function-space Parameterization of Neural Networks for Sequential Learning (ICLR 2024 · similarity 0.87)
- Tractable Function-Space Variational Inference in Bayesian Neural Networks (NeurIPS 2022 · similarity 0.87)
- Function-Space Regularization in Neural Networks: A Probabilistic Perspective (ICML 2023 · similarity 0.86)
- FSP-Laplace: Function-Space Priors for the Laplace Approximation in Bayesian Deep Learning (NeurIPS 2024 · similarity 0.85)
- Improving Adaptivity via Over-Parameterization in Sequence Models (NeurIPS 2024 · similarity 0.84)
- Thompson Sampling in Function Spaces via Neural Operators (NeurIPS 2025 · similarity 0.84)
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.
