A0260
Title: Computationally efficient SVM-based sufficient dimension reduction
Authors: Andreas Artemiou - University of Limassol (Cyprus) [presenting]
Abstract: Support-vector-machine-based (SVM-based) sufficient dimension reduction has been shown to be a great alternative to the inverse moment-based methodology introduced in the early 90s because it allows for linear and nonlinear feature extraction under a unified framework. Although it performed a more accurate estimation of the dimension reduction subspace of interest (also known as the Central Subspace), it was computationally more challenging. Although a number of projects have addressed this issue, we present an alternative approach which utilizes Twin-SVM.
