Andres Potapczynski

Bias-Free Scalable Gaussian Processes via Randomized Truncations

2021 · arXiv (Cornell University)

Scalable Gaussian Process methods are computationally attractive, yet introduce modeling biases that require rigorous study. This paper analyzes two common techniques: early truncated conjugate gradients (CG) and random Fourier features (RFF). We find that both …