Statistical Crosstalk Analysis via Probabilistic Machine Learning Surrogates

This paper discusses the application of a probabilistic surrogate modeling technique, based on Gaussian process regression (GPR), to the uncertainty quantification (UQ) of crosstalk. Compared to traditional deterministic surrogate models, the GPR provides a stochastic process that carries an estimate of the model uncertainty. This allows assigning confidence bounds to the model prediction and, in an UQ scenario, to statistical estimates. The advocated method is illustrated through its application to a literature test case.