Ghost tasking for parametrized Gaussian Processes solving linear differential equations
Ghost tasking is a systematic method to make non-parametrizable systems effectively parametrizable for Gaussian Processes (GPs) using auxiliary tasks. This technique embeds the system into a higher dimension, computes a null space, and uses it to parametrize the GP kernel. It allows constructing parametrized GPs with low task counts and latent functions, proving superior in inverse problems with limited data.