Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Pathwise Conditioning of Gaussian Processes
James T. Wilson, Viacheslav Borovitskiy, Alexander Terenin, Peter Mostowsky, Marc Peter Deisenroth
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 pathwise conditioning with random Fourier feature prior instead of location-scale (exact) prior draws | pathwise Gaussian updates · Matern-5/2 prior on R^4, 1024 test locations | runtime vs 2-Wasserstein error | Improved |
| use sparse updates in the canonical basis instead of RFF-based updates (m = n/4 basis functions) | pathwise posterior sampling · Matern-5/2 prior on R^4 | 2-Wasserstein distance | Improved |
| use sparse updates with m much smaller than n | pathwise posterior sampling · Matern-5/2 prior on R^4 | runtime | Improved |
| use conjugate-gradient solver instead of exact Cholesky for Gaussian pathwise updates | pathwise Gaussian update · Matern-5/2 prior on R^4 | accuracy and runtime | Mixed |
| precompute the inverse with CG (CG+ variant) for caching | CG pathwise update · Matern-5/2 prior on R^4 | caching efficiency and numerical error | Mixed |
| increase number of observations n towards number of Fourier features l = 1000 with unified Fourier-basis update | RFF approximate posterior · squared exponential kernel, n in {10, 100, 1000} | posterior quality (variance starvation) | Hurt |
| replace unified Fourier-basis posterior with decoupled posterior using canonical basis k(., X_n) (ablation) | RFF approximate posterior · squared exponential kernel, n close to l = 1000 | posterior quality | Improved |
Closest papers by meaning
- Efficiently sampling functions from Gaussian process posteriors (ICML 2020 · similarity 0.92)
- Conditioning Gaussian Processes on Almost Anything (NeurIPS 2026 · similarity 0.92)
- Turbocharging Gaussian Process Inference with Approximate Sketch-and-Project (NeurIPS 2025 · similarity 0.89)
- Sampling from Gaussian Process Posteriors using Stochastic Gradient Descent (NeurIPS 2023 · similarity 0.88)
- Robust and Computation-Aware Gaussian Processes (NeurIPS 2025 · similarity 0.86)
- Improving Linear System Solvers for Hyperparameter Optimisation in Iterative Gaussian Processes (NeurIPS 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.
