Alternative approach to maximum-entropy inference

摘要: A consistent approach to the inference of a probability distribution given limited number expectation values relevant variables is discussed. There are two key assumptions: that experiment can be independently repeated finite (not necessarily large) times and theoretical observables estimated from their measured sample averages. Three independent but complementary routes for deriving form these assumptions reviewed. All three lead unique which identical with one obtained by maximum-entropy formalism. The present derivation thus provides an alternative problem does not invoke Shannon's notion missing information or entropy. more in scope than proposed Jaynes, has advantage it objective operational origin "given" specified.