A. Torres Signes, M. P. Frías Bustamante, M. D. Ruiz Medina
An intrinsic dynamical multivariate functional regression from manifold-valued curve data is presented, inspired on Fréchet regression. A regularization scheme, based on Rigged-Hilbert spaces and Kuelb’s lemma, is adopted to compute Fréchet weights from the log-mapped curve regressors in the tangent space. The manifold-valued curve regressors are then selected at each time, from a set of candidates, by minimizing the Fréchet weighted sum of the supremum geodesic distances of the spherical curve response to the elements of the candidate regressor set. A simulation study is carried out to test the suitability of the model selection procedure, within a family of time-correlated spherical curves generated by applying inverse Von Mises-Fisher transform to a family of vector processes, simulated from matrix differential stochastic equations.
Keywords: Dynamical Fréchet-like spherical curve regression, manifold-valued correlated curve data, functional model selection
Scheduled
GT06 Functional Data Analysis IV. Functional Time Series
June 9, 2022 10:10 AM
Grade Hall
