Properties of the Hypothesis Space and their Effect on Machine Learning

Abstract

The best method to use for machine learning depends on the problem. This thesis considers one aspect of machine learning problems: How do the properties of the hypothesis space affect machine learning? It collects academic advances on how dimensionality and representational capacity of the space and the presence of local optima affect machine learning. Useful additions to generic machine learning methods are listed that deal with these properties. The result is a collective overview on how to design a machine learning process that uses these properties of the hypothesis space.