Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Linear cost and exponentially convergent approximation of Gaussian Matérn processes on intervals
David Bolin, Vaibhav Mehandiratta, Alexandre B. Simas
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 the order m of the rational approximation (ablation) | Rational approximation of Matern GP · Covariance error, n=5000, interval [0,50] | L2 and sup-norm covariance error: error decreases exponentially in m | Improved |
| use rational approximation instead of equal-cost competing methods | Rational approximation, m=3 and m=5 · Covariance approximation, n=5000 | covariance error | Improved |
| use rational approximation instead of FEM at calibrated mesh | FEM baseline · Covariance error, m=3 and m=5 | covariance error / stability: FEM stable only for nu < 0.9 at m=3 and nu < 0.5 at m=5 | Improved |
| use PCA as an idealized low-rank method rather than rational approximation | PCA baseline · Covariance error, large nu | covariance error | Mixed |
| compare rational approximation with nnGP for posterior mean and sd | nnGP baseline · GP regression, sigma_e = 0.1 | L2 error of posterior mean and sd: nnGP best for nu < 1; similar for nu > 1, rational slightly better for the mean | Mixed |
| compare rational approximation with nnGP at higher noise | nnGP baseline · GP regression, sigma_e = sqrt(0.1) | L2 error of posterior mean and sd: rational best for nu > 1; for nu in (0.5,1) best mean but nnGP best sd | Improved |
| use rational approximation instead of state-space, Fourier, tapering and PCA at equal cost | State-space, Fourier, Taper, PCA baselines · GP regression prediction errors | L2 error of posterior mean and sd | Improved |
| use rational approximation instead of FEM for prediction | FEM baseline · GP regression prediction errors | L2 error of posterior mean and sd: equal for nu <= 0.5, slightly worse for nu > 0.5 | Mixed |
| add prediction locations that are not nnGP support points (ablation) | nnGP · Scenario 1 and 2, joint confidence band, nu=1 | coverage error 0.9 - p and KL divergence: error small at n=0 but grows quickly with n | Hurt |
| use rational approximation instead of nnGP and FEM for process approximation | nnGP and FEM baselines · Scenario 1 and 2, nu=1, rho = 0.5, 1, 2 | coverage error and KL divergence | Improved |
| use rational approximation instead of nnGP for prior KL divergence | nnGP baseline · Scenario 1 prior KL divergence, nu=1 | KL divergence: nnGP good only when all locations are support points | Improved |
Closest papers by meaning
- Convergence of Sparse Variational Inference in Gaussian Processes Regression (JMLR 2020 · similarity 0.90)
- Posterior and Computational Uncertainty in Gaussian Processes (NeurIPS 2022 · similarity 0.87)
- Efficiently sampling functions from Gaussian process posteriors (ICML 2020 · similarity 0.87)
- Turbocharging Gaussian Process Inference with Approximate Sketch-and-Project (NeurIPS 2025 · similarity 0.86)
- Bayesian Optimization through Gaussian Cox Process Models for Spatio-temporal Data (ICLR 2024 · similarity 0.86)
- Computation-Aware Gaussian Processes: Model Selection And Linear-Time Inference (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.
