Validation design I: construction of validation designs via kernel herding
Abstract
We construct validation designs aimed at estimating the integrated squared prediction error of a given design . Our approach is based on the minimization of a maximum mean discrepancy for a particular kernel, conditional on , so that sequences of nested validation designs can be constructed incrementally by kernel herding. Numerical experiments show that key features for a good validation design are its space-filling properties, in order to fill the holes left by and properly explore the whole design space, and the suitable weighting of its points, since evaluations far from tend to overestimate the global error. A dedicated weighting method, based on a particular kernel, is proposed. Numerical simulations with random functions show the superiority the method over more traditional validation based on random designs, low-discrepancy sequences, or leave-one-out cross validation.
Cite
@article{arxiv.2112.05583,
title = {Validation design I: construction of validation designs via kernel herding},
author = {Luc Pronzato and Maria-João Rendas},
journal= {arXiv preprint arXiv:2112.05583},
year = {2021}
}