Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Learning Expressive Priors for Generalization and Uncertainty Estimation in 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 |
|---|---|---|---|
| Learned prior (from MNIST) instead of isotropic prior, without PAC-Bayes objectives (ablation) | LA posterior on NotMNIST with LeNet-5 · NotMNIST PAC-Bayes bounds | Catoni bound: 0.990 to 0.978 | Helped |
| Per-layer curvature scaling optimized via PAC-Bayes bound (ablation) | LA posterior with learned prior · NotMNIST PAC-Bayes bounds | McAllester bound: 1.347 to 1.098 | Helped |
| Frequentist projection added on top of curvature scaling (ablation) | LA posterior with learned prior · NotMNIST PAC-Bayes bounds | Catoni bound: 0.925 to 0.885 | Helped |
| Learned ImageNet prior with BPNN lateral connections instead of PNN | Continual learning on WRGBD with ResNet-50 · WRGBD continual learning | average accuracy: +0.6 over PNN wd 1e-5 | Helped |
| Learned prior vs isotropic prior around pre-trained mean in few-shot learning | ResNet-50 few-shot on eight datasets · Few-shot (up to 25 shots) | precision, ECE, calibration AUC, OOD AUC | Helped |
| Learned prior with lower temperature scaling needs (ablation) | LA on NotMNIST with LeNet-5 · Cold posterior study | accuracy vs temperature | Helped |
| Learned prior with ResNet-50 on ImageNet-1K cold posterior experiment (larger scale) (ablation) | ResNet-50 cold posterior · ImageNet-1K / CIFAR-10 | accuracy | Mixed |
| Sums-of-Kronecker-product power method vs naive sum approximation (ablation) | Approximation of sum of Kronecker products · Synthetic sums-of-Kronecker products | relative Frobenius error | Helped |
Closest papers by meaning
- Bayesian Deep Learning and a Probabilistic Perspective of Generalization (NeurIPS 2020 · similarity 0.89)
- Conservative Uncertainty Estimation By Fitting Prior Networks (ICLR 2020 · similarity 0.89)
- Deterministic PAC-Bayesian generalization bounds for deep networks via generalizing noise-resilience (ICLR 2019 · similarity 0.88)
- Modeling Uncertainty by Learning a Hierarchy of Deep Neural Connections (NeurIPS 2019 · similarity 0.88)
- Non-vacuous Generalization Bounds at the ImageNet Scale: a PAC-Bayesian Compression Approach (ICLR 2019 · similarity 0.88)
- Generalization Guarantees for Representation Learning via Data-Dependent Gaussian Mixture Priors (ICLR 2025 · similarity 0.87)
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.