Transformation-invariant kernels in GPyTorch
This package provides kernels for Gaussian processes that are invariant to group transformations, also known as symmetries.
The core class is GroupInvariantKernel, which can be composed with other GPyTorch kernels to make them invariant to a group of transformations.
Example groups are in invariant_kernels/transformation_groups.
The package was developed for the paper Sample-efficient Bayesian Optimisation Using Known Invariances, NeurIPS 2024. Code for the experiments in the paper can be found at theo-brown/bayesopt_with_invariances.