Section: New Results

Gaussian process in RKHS

Participants : Alain Celisse, Jérémie Kellner.

Since numerous papers make a Gaussian assumption for observations in the reproducing kernel Hilbert space (RKHS), it is important to be able to assess the validity of this crucial assumption. As long as it has been validated, the Gaussian framework can be further used to infer statistical properties of the population at hand (mean, variance,...).

A statistical test has been designed to address such questions at the RKHS level. It is fully computationally efficient and provides really good power in numerous settings. Theoretical properties for the test statistic have been derived as well.