Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Fast Direct Methods for Gaussian Processes
Sivaram Ambikasaran, Daniel T. Foreman-Mackey, Leslie Greengard, David W. Hogg, Michael P. O'Neil
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 |
|---|---|---|---|
| increase data dimension from 1 to 3 (Gaussian kernel) | HODLR factorization · Gaussian covariance, n=10,000 | factor time (s): 0.11 -> 0.50 -> 17.3 | Hurt |
| scale n from 10,000 to 1,000,000 (1D Gaussian) | HODLR factorization · Gaussian covariance, 1D data | factor time (s): 0.11 -> 23.2 | Improved |
| use scaled hypercube data instead of unscaled in d=8 | HODLR factorization · Gaussian covariance, n=5000, d=8 | factor time (s): 7.773 -> 138.4 | Hurt |
| use scaled hypercube data instead of unscaled in d=64 | HODLR factorization · Gaussian covariance, n=5000, d=64 | factor time (s): 0.059 -> 142.2 | Hurt |
| increase dimension beyond d=10 with fixed n (scaled data) | HODLR factorization · Gaussian covariance, scaled, n=5000 | factor time (s): 138.4 -> 138.7 | No clear effect |
| replace Gaussian by exponential kernel (1D, n=10^6) | HODLR factorization · 1D covariance, n=10^6 | factor time (s): 23.2 -> 8.05 | Improved |
| replace Gaussian by biharmonic kernel (1D, n=10^6) | HODLR factorization · 1D covariance, n=10^6 | factor time (s): 23.2 -> 81.9 | Hurt |
| vary p_max (smallest block size) at fixed epsilon (ablation) | HODLR factorization · GP regression, 1024 points | RMSE | No clear effect |
| vary factorization precision epsilon | HODLR factorization · GP regression, 1024 points | RMSE difference to exact | Mixed |
Closest papers by meaning
- High-Dimensional Gaussian Process Inference with Derivatives (ICML 2021 · similarity 0.87)
- Embarrassingly Parallel Inference for Gaussian Processes (JMLR 2019 · similarity 0.87)
- Solving High Frequency and Multi-Scale PDEs with Gaussian Processes (ICLR 2024 · similarity 0.87)
- Linear cost and exponentially convergent approximation of Gaussian Matérn processes on intervals (JMLR 2025 · similarity 0.86)
- Scalable Gaussian-process regression and variable selection using Vecchia approximations (JMLR 2022 · similarity 0.86)
- Scalable Derivative Gaussian Processes via Exact Gradient Reduction (NeurIPS 2026 · 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.
