05/01/1991 09:00 AM
Computer Science
In a one pass adaptive statistical codng it is required to have a probability distribution on the set of coding alphabet to be used for a symbol *before* it is observed. In particular at the very beginning we have to start with a probability distribution without having any data. This condition is so Bayesian in nature that it is not surprising that a number of people have used the Bayesian approach to the problem. In this report we combine various approaches to this initialization problem for adaptive binary encoding under a general framework. In fact we introduce a single family of priors and show it covers all the known methods of initializing the counts and statistics employed in binary coding. We hope that this general formula makes it easier to choose the most appropriate method for the initialization and what is called *fast attack* for particular problems at hand. We have also added an appendix that gives some historical and mathematical explanation for the most well known cases of the family of priors that we consider.
