A0724
Title: Functional predictor selection and its nonasymptotic behavior
Authors: Jun Song - Korea University (Korea, South) [presenting]
Abstract: A new method of functional prediction and estimation is presented in a scalar-on-function regression model. In particular, a functional adaptive group-lasso type penalization is applied to the regression problem when multivariate functional data are predictors of the functional regression model. Introducing a new penalty specific to infinite-dimensional functional data can relieve stringent assumptions for theoretical verification. The consistency of the method is shown, and the nonasymptotic behaviour of the method is investigated under a finite sample based on this new penalty type. Lastly, simulation and real data application show the effectiveness and validity of the method.
