Hello,

I have data sampled on a 2D regular rectangular grid, and would like to use Gaussian process methods for modelling.

My data can potentially be very large, so I’d like to use some kind of speedup or approximation. I’ve seen a few papers/articles/posts mention that regular grids allow for speedup of gaussian process calculations

- http://proceedings.mlr.press/v31/luo13b.pdf
- https://web.stanford.edu/~shenoy/GroupPublications/CunninghamShenoySahaniICML2008.pdf
- https://groups.google.com/g/stan-users/c/YGLJT1cfNUc/m/-UfvZr2nDQAJ
- Amos Storkey - Research - Toeplitz Methods and Gaussian Processes

The math is all very much over my head.

Is this method applicable in 2D input space? The link above on toeplitz matrices seems to indicate that it is not.

Are there any examples of grid based GPs in Stan that I could refer to?

If grid based speedup doesn’t work for my problem, then I can use the HSGP method instead. I’m just curious if there are any resources.

Thank you,

Scott