Is the Maximum Entropy Principle Operationally Justifiable?
- 389 Downloads
Let X be a random variable originally believed to have distribution Q. When new information is obtained suggesting that the distribution of X actually belongs to a set of distributions Π not containing the original guess Q, this should be updated to conform with the new information. Intuitively a proper updating should be that element of Π which is closest to the original guess Q. It remains to specify the measure of distance between distributions to be used to find this closest element.
KeywordsMaximum Entropy Common Distribution Close Element Common Range Conditional Limit
- T. Cover, B.S. Choi, and I. Csiszár, “Conditional Limit Theorems under Markov Conditioning,” to appear IEEE Trans. Inf. Theory.Google Scholar