Introduction to Information Theory (67548)
Spring 2006
Hours and Location
Lecture: Mondays, 10:00 - 11:45, Shprintzak 115
Exercise: Mondays, 16:00 - 17:45, Shprintzak 28
Staff
Teacher:
Prof. Naftali Tishby
Reception hour: Sundays, 14:00-15:00, Ross 214
Teaching Assistant:
Talya Meltzer infot@cs.huji.ac.il
Reception hour: Mondays, 18:00-19:00, Ross 67
Syllabus
- Shannon's communication model - the fundamental problems: sources
(compression) and channels (error correction).
- Measures of information: the uniqueness of Entropy and Mutual
Information, chain rules, Jensen & data processing inequalities,
Fano inequality.
- Asymptotic properties of Entropy, relative Entropy and
Information, AEP, Entropy rate and Shannon-McMillen Theorem for
stationary stochastic processes.
- Lossless data compression. Codes, Kraft inequality, optimal
codes, Shannon's first theorem, Huffman codes and questioners.
- Universal compression, LZ code, Kolmogorov Complexity and
algorithmic randomness.
- Joint typicality and AEP for mutual information, differential
entropy.
- Maximum Entropy and Minimum Mutual Information as Inference
methods.
- Channel Capacity (with and without cost constraint). BSC, the
Gaussian channel and the C(E) function. Block coding and Simple
examples of error correcting codes.
- Lossy compression: Rate Distortion Theory.
- Joint source-channel coding - and optimal cost-distortion
tradeoff. MAYBE: Optimal single letter codes.
- Information Theory, Statistics and Learning.
- Open problems. Connections with other fields (control theory,
economy & game-theory, etc.)
Requirements
You have to submit at least 75% of the exercises.
The exercises grade would be 25% of your final grade.
Course Administration