Title: Dimension reduction through unsupervised learning on predictors
Authors: Jae Keun Yoo - Ewha Womans University (Korea, South) [presenting]
Abstract: A methodology is proposed to estimate the central subspace in sufficient dimension reduction context by utilizing unsupervised learning on a set of predictors. By using the additional information on the predictors, the response is sliced within each class of the predictors. The potential advantage of this double slicing scheme is no requirements of linearity and constant variance conditions in methodological developments, which are normally assumed in most sufficient dimension reduction methodologies. Also, the central subspaces can be exhaustively estimated under mild condition. Numerical studies confirm the theories, and real data analyses are presented.