B0750
Title: Generalized tensor eigenpairs for moment-based projection pursuit
Authors: Nicola Loperfido - University of Urbino (Italy) [presenting]
Abstract: Projection pursuit is a multivariate statistical technique aimed at finding interesting data projections. The projection index measures the interestingness of a given data projection. In moment-based projection pursuit, the projection index depends on some moments of the projection, for example, its skewness or its kurtosis. Generalized tensor eigenpairs are introduced as natural tools for moment-based projection pursuit since they lead to the optimization of a multilinear form under a constraint on another multilinear form in the same variables. They include tensor eigenvectors and generalized matrix eigenvectors as special cases. The use of generalized tensor eigenpairs is investigated when addressing the problems posed by high-dimensional data and multivariate outliers. Some connections are also highlighted with independent component analysis and invariant coordinate selection.
