ON ENSEMBLE LEARNING VIA PROJECTION PURSUIT

Abstract

The technique of projection is both popular and powerful in statistics, machine learning and pattern recognition, which is a helpful tool to study multivariate dimensional data. Specifically, instead of using the original data directly in the analysis this technique finds the lower-dimension subspace which can capture the data structure as much as possible. Therefore, projection technique is an efficient way to analyze multivariate dimensional data since the above mentioned dimension reduction helps us dramatically relieve computation burdens. In this thesis, we study two ensemble learning algorithms, which aim to estimate a conditional mean function, based on the technique of projection.