Abstract
The universal coding of Zipf distributions are discussed. Coding schemes whose redundancy increases slower than n are known as universal. The universal coding scheme can be considered as an algorithm for combining expert advice whose code length is equal to the cumulative log loss and the redundancy is the difference between the loss of the combining algorithm and the loss of the best expert. When compressing natural-language text, it is reasonable to code the text a word at a time, thereby relying on the distribution of the words. One approach to reduce the redundancy in that case would be to restrict the collection of possible distributions over the words.
Original language | English (US) |
---|---|
Pages (from-to) | 736-737 |
Number of pages | 2 |
Journal | Lecture Notes in Artificial Intelligence (Subseries of Lecture Notes in Computer Science) |
Volume | 2777 |
DOIs | |
State | Published - 2003 |
Externally published | Yes |
Event | 16th Annual Conference on Learning Theory and 7th Kernel Workshop, COLT/Kernel 2003 - Washington, DC, United States Duration: Aug 24 2003 → Aug 27 2003 |
ASJC Scopus subject areas
- Theoretical Computer Science
- Computer Science(all)