Statistical Inference through Data Compression
Rudi Cilibrasi · Data Archiving and Networked Services (DANS) · 2006
ordinary minds to carry out.Heuristics have even become associated with inevitable cognitive illusions and irrationality.This author sides with Goldstein and Gigerenzer in the view that sometimes "less is more"; the very fact that things are unknown to the naive observer can sometimes work to his advantage.The recognition heuristic is an important, reliable, and conservative general strategy for inductive inference.In a similar vein, the NCD based techniques shown in this thesis provide a general framework for inductive inference that is robust against a wide variety of circumstances. Contents of this ThesisIn this chapter a summary is provided for the remainder of the thesis as well as some historical context.In Chapter 2, an introduction to the technical details and terminology surrounding the methods is given.In chapter 3 we introduce the Normalized Compression Distance (NCD), the core mathematical formula that makes all of these experiments possible, and we establish connections between NCD and other well-known mathematical formulas.In Chapter 4 a tree search system is explained based on groups of four objects at a time, the so-called quartet method.In Chapter 5 we combine NCD with other machine learning techniques such as Support Vector Machines.In Chapter 6, we provide a wealth of examples of this technology in action.All experiments in this thesis were done using the CompLearn Toolkit, an open-source general purpose data mining toolkit available for download from the http://complearn.org/website.In Chapter 7, we show how to connect the internet to NCD using the Google search engine, thus providing the advanced sort of subjective analysis as shown in Figure 1.2.In Chapter 8 we use these techniques and others to trace the evolution of the legend of Saint Henry.In Chapter 9 we compare CompLearn against another older tree search software system called PHYLIP.Chapter 10 gives a snapshot of the online documentation for the CompLearn system.After this, a Dutch language summary is provided as well as a bibliography, index, and list of papers by R. Cilibrasi.