Sparse kernel extreme components analysis

The extreme components analysis is the naturally generalization of principal components analysis and minor components analysis, so it also suffers the same difficulties that principal components analysis encountered for nonlinear projection of data. Inspired by M. E. Tipping' idea, we introduce the sparse kernel extreme components analysis for nonlinear projection of data in this paper. Approximating the covariance matrix in feature space, we may obtain a highly sparse form of kernel extreme components analysis with negligible loss of performance. Afterwards, multi-layer support vector machine is presented based on sparse kernel extreme components analysis for multi-input multi-output regression estimation problem. The result of numerical simulation demonstrates that multi-layer support vector machine is a very effective method for multi-input multi-output regression estimation problem and shows good approximate performance.