Jacob (Xuankang) Zhu
Title: Efficient learning algorithms for classification of data by a nonlinear decision boundary
Date: August 24th, 2022
Time: 1:00PM PST
Location: Remote delivery - Zoom
Abstract
In this thesis, we study the classification of data by a nonlinear classification boundary in the case the decision boundary is determined by the inverse of a given function. We derive efficient machine learning algorithms by building models on the Gabriel Edited Set, which is an approximate decision-boundary consistent reduced set. The algorithms account for the geometry of the equivalence classes determined by the inverse of the function. We numerically investigate the behaviour and accuracy of the algorithm on multiple simulated examples.
Keywords: Classification; Inverse problem; proximity graph; Gabriel edited set; localized SVM