Compressed Vector-Valued Kernel Ridge Regression for Complex-Valued Parametric Modeling of High-Speed Interconnects

This paper investigates a compressed formulation of Vector-Valued Kernel Ridge Regression (VV-KRR) for complexvalued modeling. The method is applied to the parametric modeling of the frequency response of a high-speed interconnect, as a function of four design parameters. Its effectiveness is assessed in terms of prediction accuracy and model compactness and compared with an uncompressed formulation.