Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Exploring the Uncertainty Properties of Neural Networks’ Implicit Priors in the Infinite-Width Limit
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 |
|---|---|---|---|
| Replace finite-width CNN classification with softmax NNGP classification | CNN · CIFAR10 | ECE: 0.283 -> 0.031 | Helped |
| Replace finite-width CNN classification with softmax NNGP classification | CNN · CIFAR10 | Accuracy: 0.576 -> 0.609 | Helped |
| Replace finite-width CNN classification with softmax NNGP classification | CNN · Fog 5 | ECE: 0.420 -> 0.134 | Helped |
| Replace finite-width CNN classification with softmax NNGP classification | CNN · Fog 5 | Accuracy: 0.287 -> 0.289 | No clear effect |
| Use a convolutional rather than fully connected NNGP kernel for classification | NNGP-C · CIFAR10 | ECE: 0.072 -> 0.031 | Helped |
| Replace an EfficientNet-B3 last-layer baseline with an NNGP head | EfficientNet-B3 last-layer methods · CIFAR10-C | ECE (50th): 0.028 -> 0.025 | Helped |
| Replace an EfficientNet-B3 last-layer baseline with an NNGP head | EfficientNet-B3 last-layer methods · CIFAR10-C | Accuracy (50th): 0.813 -> 0.813 | No clear effect |
| Use NNGP regression instead of an RBF GP | UCI regression · Power Plant | RMSE: 3.82 ± 0.04 -> 3.61 ± 0.04 | Helped |
| Use NNGP regression instead of an RBF GP | UCI regression · Yacht Hydrodynamics | RMSE: 0.60 ± 0.07 -> 0.41 ± 0.04 | Helped |
| Fine-tune the embedding while retaining an NNGP final layer (ablation) | NNGP-LL · CIFAR10-C | Accuracy (50th): 0.894 -> 0.896 | Helped |
| Fine-tune the embedding while retaining an NNGP final layer (ablation) | NNGP-LL · CIFAR10-C | NLL (50th): 0.320 -> 0.374 | Hurt |
| Use the exact or pairwise posterior-variance confidence heuristic instead of softmax of the posterior mean (ablation) | C-NNGP-R · CIFAR10-C | Calibration | Helped |
Closest papers by meaning
- Critical feature learning in deep neural networks (ICML 2024 · similarity 0.89)
- Uncertainty Quantification with the Empirical Neural Tangent Kernel (NeurIPS 2025 · similarity 0.87)
- A Study of Bayesian Neural Network Surrogates for Bayesian Optimization (ICLR 2024 · similarity 0.87)
- Bayesian Neural Scaling Law Extrapolation with Prior-Data Fitted Networks (ICML 2025 · similarity 0.87)
- An Infinite-Width Analysis on the Jacobian-Regularised Training of a Neural Network (ICML 2024 · similarity 0.86)
- Deep Kernel Posterior Learning under Infinite Variance Prior Weights (ICLR 2025 · similarity 0.86)
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.