Decoupling multivariate functions using a non-parametric Filtered CPD approach

Abstract Black-box model structures are dominated by large multivariate functions. Usually a generic basis function expansion is used, e.g. polynomial basis, and the parameters of tuned given data. This pragmatic often necessary step considering black-box nature problem. However, having identified suitable function, there no need to stick original basis. So-called decoupling techniques aim at translating functions into an alternative thereby both reducing number retrieving underlying structure. In this work filtered canonical polyadic decomposition (CPD) introduced. It non-parametric method which able retrieve decoupled even when facing non-unique decompositions. Tackling obstacle paves way for modelling applications.