Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Gaussian Process Behaviour in Wide Deep 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 |
|---|---|---|---|
| Increase network width | Random ReLU fully connected networks · 10 standard normal inputs in 4 dimensions; varying depth and width functions | squared MMD to the corresponding Gaussian process | Helped |
| Increase the number of hidden layers at the same width | Random ReLU fully connected networks · 10 standard normal inputs in 4 dimensions | squared MMD to the corresponding Gaussian process | Hurt |
| Use largest-first or largest-last widths instead of identity widths | Random ReLU fully connected networks · 10 standard normal inputs in 4 dimensions | squared MMD to the corresponding Gaussian process | Helped |
| Use largest-last rather than largest-first widths | Random ReLU fully connected networks · 10 standard normal inputs in 4 dimensions | squared MMD to the corresponding Gaussian process | Mixed |
| Use a finite Bayesian deep network instead of its Gaussian process analogue | Bayesian deep network with 3 hidden layers and 50 units per layer · Five of the six posterior-comparison datasets | agreement of posterior quantities | No clear effect |
| Use a finite Bayesian deep network instead of its Gaussian process analogue | Bayesian deep network with 3 hidden layers and 50 units per layer · Yacht | test hold-out log likelihood | Helped |
Closest papers by meaning
- Wide Feedforward or Recurrent Neural Networks of Any Architecture are Gaussian Processes (NeurIPS 2019 · similarity 0.92)
- Deep Kernel Posterior Learning under Infinite Variance Prior Weights (ICLR 2025 · similarity 0.87)
- Beyond IID weights: sparse and low-rank deep Neural Networks are also Gaussian Processes (ICLR 2024 · similarity 0.86)
- Quantitative convergence of trained neural networks to Gaussian processes (NeurIPS 2025 · similarity 0.86)
- Convolutional Deep Kernel Machines (ICLR 2024 · similarity 0.86)
- Critical feature learning in deep neural networks (ICML 2024 · 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.