A variant of IIS algorithm

Interestingly, the principle of maximum entropy seems to be different from other machine learning approaches. A couple of more accurate and predictive models, such as neural nets, hidden Markov model continue to rapidly develop, which do help people make more and more detailed prediction, but these accomplishments cannot keep pace with the requirements that people make for the tasks of machine learning. Ironically, in some sense maximum entropy is based on the premise that we know as little information as possible, once the given constraint is satisfied. Maximum entropy model is intuitively appealing for its simplicity to implement, in addition to ability to incorporate distinct classes of feature without any performance impairment and suffer from independent assumption.