Features:

Many methods for compressing data sequences assume that the entropy of the sequence is known. However, for sequences produced by nature (including humans), the entropy is often unknown and has to be estimated.

There exist various entropy estimation techniques, for discrete and continuous sources, with and without memory, as well as efficient algorithms for their implementation.

In this project, the student will study several such techniques and algorithms and implement some of them. One can then measure the performance of the algorithm using test sequences, but also apply the algorithm to estimate the entropy of natural data.

Objective:

Study various entropy estimation algorithms, implement and test them for various sources. Sources might include artificially produced test sequences, but also natural sequences like text or DNA/protein sequences from biology.

Prerequisites:

• Solid background in probability
• information theory
• good programming skills (preferably in C)

Benefit

You will learn various algorithms for entropy estimation and enlarge your programming experience.

References:

Contact:

Etienne Perron (LICOS) * Email: etienne.perron@epfl.ch * Office: INR-033 * Tel: 36457

Supervisor: Prof. Suhas Diggavi

back to master semester projects