Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Variational Fourier Features for Gaussian Processes
James Hensman, Nicolas Durrande, Arno Solin
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 variational Fourier features instead of Random Fourier Features | Matern-1/2 GP regression · 1D toy regression | approximation quality vs full GP | Improved |
| increase number of variational features M (ablation) | VFF · 1D toy regression | ELBO tightness | Improved |
| set interval [a,b] too narrow, within the data region (ablation) | VFF · Matern-3/2 toy regression (true marginal likelihood -15.99) | ELBO | Hurt |
| set interval [a,b] too large (ablation) | VFF · Matern-3/2 toy regression | approximation quality with M=8 or 16 | Hurt |
| use VFF (Kronecker) instead of sparse GP with inducing inputs | sparse GP · product-kernel toy data, d=2..4 | wall-clock time at same number of inducing variables | Improved |
| use VFF (Kronecker) instead of sparse GP with inducing inputs | sparse GP · product-kernel toy data, d=2..4 | KL at same number of inducing variables | Hurt |
| use VFF instead of sparse GP, matched computational time | sparse GP · product-kernel toy data | KL divergence vs time | No clear effect |
| use VFF instead of SVIGP | SVIGP · Airline delays (N=5,929,413) | MSE: 0.791 -> 0.827 | Hurt |
| use VFF instead of SVIGP | SVIGP · Airline delays (N=100,000) | MSE: 0.79 -> 0.82 | Hurt |
| use VFF instead of String GP | String GP · Airline delays (N=5,929,413) | MSE: 0.90 -> 0.827 | Improved |
| use VFF instead of rBCM | rBCM · Airline delays (N=100,000) | MSE: 1.04 -> 0.82 | Improved |
| use VFF instead of full additive GP | Full-additive · Airline delays (N=10,000) | MSE: 0.89 -> 0.89 | No clear effect |
Closest papers by meaning
- Sparse Gaussian Processes with Spherical Harmonic Features (ICML 2020 · similarity 0.92)
- Scalable Variational Gaussian Processes via Harmonic Kernel Decomposition (ICML 2021 · similarity 0.91)
- Variational Inference for Gaussian Process Models with Linear Complexity (NeurIPS 2017 · similarity 0.91)
- New Bounds for Sparse Variational Gaussian Processes (ICML 2025 · similarity 0.89)
- Sparse Gaussian Processes: Structured Approximations and Power-EP Revisited (NeurIPS 2025 · similarity 0.88)
- Solving High Frequency and Multi-Scale PDEs with Gaussian Processes (ICLR 2024 · similarity 0.88)
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.
