Approximate Inference in Graphical Models (PhD Thesis)
Approximate Inference in Graphical Models [pdf]
Philipp Hennig
University of Cambridge, 2011
My PhD thesis was prepared from 2007 to 2010 under the supervision of David J C MacKay, at the Cavendish Laboratory and also under the supervision of Thore Graepel, of Microsoft Research. It was generously funded through a Microsoft PhD Scholarship. The thesis was submitted to the University of Cambridge in November 2010, and defended successfully on 11 January 2011.
The thesis is a heterogeneous series of applied inference algorithms. Although the studied applications are very diverse (they include inference on optimal play in games, inference of personality traits from psychometric questionnaires, and fast inference of the topics of documents from the documents' words and metadata), a joining thread is provided by the use of approximate inference methods and graphical models. The thesis also contains a compact introduction into the field of graphical model inference.