Nonlinear function decomposition

While linear functions are widely used and well-understood, for nonlinear multivariate vector functions it is unclear how to i) define their complexity, ii) reduce the complexity and iii) increase their interpretability.To solve these problems, we propose a decomposition of nonlinear functions, which can be viewed as a generalization of the singular value decomposition. We use tensor methods to compute these decompositions and apply them in the domain of nonlinear system identification.

Keywords:low-rank approximation, tensor, lage-rang benadering

Disciplines:Linear and multilinear algebra, matrix theory, Data mining, Modelling and simulation, Numerical computation