An informativeness score for optimal mixed datasets using Gaussian process regression

For many systems composed of numerous subsystems, it is useful to have predictive models of the subsystems themselves. Gaussian process regression (GPR) is a machine learning technique which gives uncertainty in its predictions. The ability to calculate prediction uncertainty leads to the ability to calculate an optimally informative training data subset. Our previous work details a process for generating subsystem models with data from a shared source of system-level training data using GPR. In this study, we present a method for calculating an optimally informative dataset for numerous subsystem models. We then demonstrate our technique’s effectiveness using a mixture of non-expert produced whole-system data and expert produced subsystem-specific data. We show that regardless of dataset size, models made with a dataset selected according to our optimally whole-system informative criterion are more accurate than models made with arbitrarily included data or datasets optimized without a whole-system view. In many cases, using optimally informative data shared across all subsystem models lead to better accuracy than randomly selected datasets that are 50% larger.

Abstract licence: CC BY

26 more references, whose wording the paper’s licence keeps private.

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.