Kernelized Normalizing Constant Estimation: Bridging Bayesian Quadrature and Bayesian Optimization

Xu caiが第一著者

In this paper, a normalizing constant on RKHS is considered as follows:

$$Z(f) = \int_{D} e^{- \lambda x} dx, \quad \lambda > 0$$

This method considers the lower bound and the upper bound of

$$f \in RKHS$$

This method considers the noiseless setting and the noisy setting for the lower bound of f, respectively.

Applicable to a multi-layer perception, a point spread function.
So authors conducted experiments with various f.

Then consider the results with these experiments.