Comparative study of Machine Learning methods for variability analysis in High-speed link

Abstract

Non-intrusive stochastic analysis of a complex system requires a fast deterministic solver to simulate the mapping between the input and output. Different machine learning methods, namely Partial Least Square regression, Gaussian Process, and Polynomial Chaos expansion can be used to represent the input - output mapping. Once they are trained to learn the mapping, they are used to replace the expensive process that generates the output given an input, such as a full-wave electrogmanetic solver. Aforementioned methods are compared in this paper when trained on a simple high-speed link.