Local to global geometric methods in information theory

This thesis treats several information theoretic problems with a unified geometric approach. The development of this approach was motivated by the challenges encountered while working on these problems, and in turn, the testing of the initial tools to these problems suggested numerous refinements and improvements on the geometric methods. In ergodic probabilistic settings, Sanov's theorem gives asymptotic estimates on the probabilities of very rare events. The theorem also characterizes the exponential decay of the probabilities, as the sample size grows, and the exponential rate is given by the minimization of a certain divergence expression. In his seminal paper, A Mathematical Theory of Communication, Shannon introduced two influential ideas to simplify the complex task of evaluating the performance of a coding scheme: the asymptotic perspective (in the number of channel uses) and the random coding argument. In this setting, Sanov's theorem can be used to analyze ergodic information theoretic problems, and the performance of a coding scheme can be estimated by expressions involving the divergence. One would then like to use a geometric intuition to solve these problems, but the divergence is not a distance and our naive geometric intuition may lead to incorrect conclusions. In information geometry, a specific differential geometric structure is introduced by means of "dual affine connections". The approach we take in this thesis is slightly different and is based on introducing additional asymptotic regimes to analyze the divergence expressions. The following two properties play an important role. The divergence may not be a distance, but locally (i.e., when its arguments are "close to each other"), the divergence behaves like a squared distance.

Department dc:contributor.department

Massachusetts Institute of Technology. Dept. of Electrical Engineering and Computer Science.

Grantor dc:publisher

Massachusetts Institute of Technology

Year dc:date.issued

2008

Author dc:creator

• Abbe, Emmanuel Auguste

Advisor dc:contributor.advisor

• Lizhong Zheng and Emre Telatar.

• Electrical Engineering and Computer Science.

Statement dc:rights

• M.I.T. theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission. See provided URL for inquiries about permission.

Licence dc:rights.uri

• http://dspace.mit.edu/handle/1721.1/7582

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/1721.1/44404

OAI identifier oai:identifier

oai:dspace.mit.edu:1721.1/44404