Supervised learning & prediction, using Gaussian Processes
Dalia Chakrabarty · 2025
In the first chapter, we came across the suggestion that it is extremely difficult to build a high-dimensional function, by sequentially designing individual component functions, while ensuring adherence of each such component, to multiple (probabilistic) inter-connected constraints. An alternative to such an approach, was proposed in collecting realisations of the sought function inside a “bag” of functions, where it is the responsibility of said bag, to transfer the correlation structure of the data, onto the inter-component correlation (i.e. correlation amongst component functions that constitute the sought high-dimensional function), and the inter-realisation correlation (i.e. correlation amongst possible realisations of this function).