Parallel Gaussian process with kernel approximation in CUDA

CUDA implementation of Gaussian process with approximated kernel. The paper linked to this repo is available as a pre-print here. The mathematical formulation was originally presented in V. Joukov and D. Kulić (2022) (pre-print available here).

Requirements

• gcc v9.4.0
• CUDA toolkit v12.2
• CMake >= v3.18
• Eigen3 v3.4
• Boost >= v1.71
• Matlab/Octave The code is tested in Ubuntu 20.04.6 LTS. Figures are generated in Matlab R2022a. Octave 5.2.0 is also supported.

Building the executables

In the desired folder, clone this repository:

git clone [email protected]:DavideCarminati/cuFAGP.git

Then cd in the folder and build the executables with CMake:

mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make

Executables are located in the build folder.

Run the example script

Once executables are built, run in the terminal:

cd ..
source example.sh

to replicate the results in the paper.

How to cite

@misc{carminati2024parallel,
      title={Parallel Gaussian process with kernel approximation in CUDA}, 
      author={Davide Carminati},
      year={2024},
      eprint={2403.12797},
      archivePrefix={arXiv},
      primaryClass={cs.DC}
}