Evaluating the squared-exponential covariance function in Gaussian processes with integral observations

This addresses a computational bottleneck for researchers and practitioners using Gaussian processes with integral observations, though it is an incremental improvement over existing methods.

The paper tackled the computational challenge of evaluating double line integrals of the squared-exponential covariance function in Gaussian processes with integral observations by proposing a method that reduces the double integral to a single integral using the error function and computes it with efficient numerical techniques. The results demonstrated superior numerical robustness and accuracy per computation time compared to existing state-of-the-art methods.