Clustering by Mutual Information

Dipl.Ing. Georg Pichler

Institute of Telecommunications
Technische Universität Wien / TU Wien, Vienna

Abstract:

Georg Pichler will present his PhD thesis on problems in multi-terminal source coding. It focuses on a setup motivated by biclustering algorithms. We analyze the resulting achievable region, provide bounds, and discuss connections of this setup with several other problems in network information theory.

In the case of a doubly symmetric binary source, we argue that there is a gap between the resulting inner and outer bounds, disproving a conjecture by Westover and O'Sullivan (2008). Furthermore, we present the proof of the two-function case of the Kumar-Courtade (2014) conjecture on the mutual information between Boolean functions.

We also extend the setup to an arbitrary number of sources, where we show tight inner and outer bounds for a multiple description variant of the CEO problem.

Biography:Biography:

Georg Pichler received his BSc degree in Electrical Engineering in 2011 from TU Wien, Vienna, Austria, and his Dipl-Ing. (MSc) degree in Telecommunications in 2013, also from TU Wien. Since 2013 he is with the Communication Theory group at the Institute of Telecommunications, TU Wien as a research assistant, working towards his PhD, which he is scheduled to complete in November 2017. He was working on multi-terminal source coding problems under the supervision of Prof. Gerald Matz and Prof. Pablo Piantanida.