Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Function Space Bayesian Pseudocoreset for Bayesian Neural Networks
Balhae Kim, Hyungi Lee, Juho Lee
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 sample (empirical) covariance instead of isotropic Gaussian for function-space posterior (ablation) | FBPC · CIFAR10 ipc 10 | test accuracy (%): 61.43 -> 62.33 | Improved |
| use sample covariance instead of isotropic Gaussian (ablation) | FBPC · CIFAR10 ipc 1 | test accuracy (%): 32.00 -> 35.45 | Improved |
| use sample covariance instead of isotropic Gaussian (ablation) | FBPC · CIFAR10 ipc 50 | test accuracy (%): 71.33 -> 71.23 | No clear effect |
| match posteriors in function space instead of weight space (BPC-fKL) | FBPC · CIFAR10 ipc 10 | test accuracy (%): 56.19 -> 62.33 | Improved |
| function space instead of weight space | FBPC · CIFAR100 ipc 10 | test accuracy (%): 28.1 -> 39.7 | Improved |
| function space instead of weight space | FBPC · Tiny-ImageNet ipc 10 | test accuracy (%): 11.4 -> 19.42 | Improved |
| function space instead of weight space | FBPC · CIFAR10-C average | test accuracy (%): 34.3 -> 47.5 | Improved |
| function space instead of weight space | FBPC · CIFAR10-C average | degradation score: 38.9 -> 23.7 | Improved |
| train pseudocoreset on multiple architectures (varying normalization) instead of one (ablation) | FBPC · CIFAR10 across normalization layers | test accuracy | Improved |
| train ResNet18 pseudocoreset jointly with ConvNet (ablation) | FBPC · CIFAR10 ipc 10, ResNet18 | test accuracy (%): 50.00 -> 54.90 | Improved |
| remove DSA augmentation (ablation) | FBPC · CIFAR10 ipc 10 | test accuracy (%): 62.33 -> 55.07 | Hurt |
| remove DSA augmentation (ablation) | BPC-fKL · CIFAR10 ipc 1 | test accuracy (%): 34.50 -> 37.26 | Mixed |
Closest papers by meaning
- On Divergence Measures for Bayesian Pseudocoresets (NeurIPS 2022 · similarity 0.91)
- Variational Bayesian Pseudo-Coreset (ICLR 2025 · similarity 0.88)
- Bayesian Pseudocoresets (NeurIPS 2020 · similarity 0.88)
- FSP-Laplace: Function-Space Priors for the Laplace Approximation in Bayesian Deep Learning (NeurIPS 2024 · similarity 0.87)
- Function-space Parameterization of Neural Networks for Sequential Learning (ICLR 2024 · similarity 0.86)
- A Study of Bayesian Neural Network Surrogates for Bayesian Optimization (ICLR 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.
