A Stochastic Active Learning Strategy for Gaussian Process Models with Application to the Uncertainty Quantification of Signal Integrity

This paper introduces an effective active learning strategy to iteratively refine the training of Gaussian process models with application to uncertainty quantification. Compared to traditional deterministic approaches or naive random sampling, the proposed approach uses a stochastic measure, i.e., the predictive variance of the output mean, to drive the acquisition of additional training points. The advocated algorithm is shown to outperform alternative strategies in the uncertainty quantification of the insertion loss of a microstrip transmission line with a discontinuity in the ground plane.