A type generator model for adaptive lossy compression
Ram Zamir, Kenneth H. Rose · 2002
We propose a random coding model which gives insight into the underlying mechanism of lossy string-matching algorithms. This idealised model is based on manipulating types rather than codewords. We model the adaptive codebook as a mixed type random code, whose type distribution evolves with time as its dimension increases. The model behavior exhibits a mechanism of natural type selection which reinforces "good types".