English

Graph Random Features for Scalable Gaussian Processes

Machine Learning 2025-09-29 v2

Abstract

We study the application of graph random features (GRFs) - a recently introduced stochastic estimator of graph node kernels - to scalable Gaussian processes on discrete input spaces. We prove that (under mild assumptions) Bayesian inference with GRFs enjoys O(N3/2)O(N^{3/2}) time complexity with respect to the number of nodes NN, compared to O(N3)O(N^3) for exact kernels. Substantial wall-clock speedups and memory savings unlock Bayesian optimisation on graphs with over 10610^6 nodes on a single computer chip, whilst preserving competitive performance.

Keywords

Cite

@article{arxiv.2509.03691,
  title  = {Graph Random Features for Scalable Gaussian Processes},
  author = {Matthew Zhang and Jihao Andreas Lin and Krzysztof Choromanski and Adrian Weller and Richard E. Turner and Isaac Reid},
  journal= {arXiv preprint arXiv:2509.03691},
  year   = {2025}
}
R2 v1 2026-07-01T05:19:59.283Z