Is Gaussian process regression function (fitrgp) compatible with GPU processing?

I am using a workstation PC with a QUADRO K6000 GPU and Xeon CPU. Currently, I need to use Gaussian process regression optimise toolbox to optimise the hyperparameters, and I am looking for a way to increase processing speed by using the GPU.

Answer 1

[Sai Kulkarni]

Hi Philip,

As per the following documentation, fitrgp does not support gpuArray arguments.

https://www.mathworks.com/help/stats/statistics-and-machine-learning-toolbox-functions-with-gpuarray-arguments.html

For more information, you can refer to the following documentation for parallel computing and GPU support across all toolboxes: https://www.mathworks.com/products/parallel-computing/parallel-support.html

[Robert]

Would it be possible to initialize a feature request on this? Regarding the matrix inversions, determinant calculations etc. involved in the (hyperparameter) optimization it looks promising to me at first sight that there could be a substantial speed-up. Thank you!