After passing the assessment of this course a student:
- knows Shannon entropy and mutual information, is able to perform computations involving these information measures and is able to estimate these measures from a data set,
- knows the Huffman, Lempel-Ziv and CTW data compression algorithms, and understands the limits on data compression,
- understands the connection between machine learning and data compression and is able to quantify performance limits on machine learning algorithms,
- is able to quantify the optimal performance that can be expected from a classification system in terms of information measures,
- knows how linear block codes can be used to achieve reliable communication over a noisy channel, understands en-/de-coding of a graph-based code and understands the limits on reliable communication.
Information theory is a mathematical theory dealing with the fundamental principles of storing, processing and transmitting data. The first half of this course covers the core concepts of information theory, including entropy and mutual information. These are then used to derive fundamental limits on data compression and communication. The second half of this course focusses on applications of information theory to statistics and machine learning. In particular, information theory will be used to develop performance limits on machine learning algorithms.