Categorization and Machine Learning : The Modeling of Human Understanding in Computers

Machine learning is the attempt to imitate human categorization of perceived reality in computers. It is driven by the desire to provide machines that are as open-minded, intelligent and flexible as humans. The central goal is to provide classifications for arbitrary types of input data: Labels that characterize the data correctly, given some examples. Machine learning has been a research topic of computer science for several decades. This book summarizes the major findings, explains the practically relevant methods and discusses their communalities and differences. In the first of three parts, we introduce the setting, goals and all necessary tools for the definition, application and evaluation of learning algorithms. The second part discusses and compares the various algorithms employed in machine categorization today. We structure them in four groups: the optimization algorithms, risk minimization approaches, those that employ probabilistic inference and those that imitate neural inference processes. Outstanding examples from the list of algorithms are the vector space mode, the support vector machine, Bayes and Markov processes, conditional random fields, radial basis function networks and methods employed for deep learning such as the Boltzmann machine. The third part reviews the algorithms and explores the theoretical frontiers of machine learning. In summary, we endeavor to provide a comprehensive yet intuitive introduction into the field of categorization. Neither parallels to human cognition are neglected nor recent developments in algorithm design or theoretical justification. As a research field, machine learning is gaining more and more attention. This book explains what it is, where it can be applied and how it is done.