Abstract

Information theory is widely used for analyzing a wide range of scientific and engineering problems, including cryptography, neurobiology, quantum computing, plagiarism detection and other forms of data analysis. Despite the safety-critical nature of some of these applications, most of the information-theoretic analysis is done using informal techniques, mainly computer simulation and paper-and-pencil analysis, and thus cannot be completely relied upon. The unreliable nature of the produced results poses a serious problem in safety-critical applications and may result in heavy financial losses or even the loss of human life. In order to overcome the inaccuracy limitations of these techniques, this thesis proposes to conduct the analysis within the trusted kernel of a higher-order-logic (HOL) theorem prover. For this purpose, we provide HOL formalizations of the fundamental theories of measure, Lebesgue integration and probability and use them to formalize some of the most widely used information-theoretic principles. We use the Kullback-Leibler divergence as a unified measure of information which is in turn used to define the main measures of information like the Shannon entropy, mutual information and conditional mutual information. Furthermore, we introduce two new measures of information leakage, namely the information leakage degree and the conditional information leakage degree and compare them with existing measures. We illustrate the usefulness of the proposed framework by tackling various applications including the performance analysis of a communication encoder used in the proof of the Shannon source coding theorem, the quantitative analysis of privacy properties of a digital communications mixer and the one-time pad encryption system using information-theoretic measures.