On the Correspondence between Gaussian Processes and Geometric Harmonics
This work connects disparate research communities, offering incremental improvements in uncertainty estimation and optimization efficiency for machine learning applications.
The paper explores the correspondence between Gaussian process regression and Geometric Harmonics, two kernel-based methods used in different contexts, showing that combining results from both can provide alternative interpretations of uncertainty and accelerate Bayesian Optimization through dimensionality reduction.
We discuss the correspondence between Gaussian process regression and Geometric Harmonics, two similar kernel-based methods that are typically used in different contexts. Research communities surrounding the two concepts often pursue different goals. Results from both camps can be successfully combined, providing alternative interpretations of uncertainty in terms of error estimation, or leading towards accelerated Bayesian Optimization due to dimensionality reduction.