The principle of maximum entropy consists when one wants to represent an imperfect knowledge of a phenomenon by a law of probability, with:

• to identify the constraints to which this distribution must answer (average, etc;)

• to choose of all the distributions answering these constraints that having largest Entropy within the meaning of Shannon.

This choice does not have anything arbitrary: of all these distributions, it is - by definition of the entropy - that of maximum entropy which contains less information , and it is thus for this reason less Arbitraire of all those which one could use.

The probability distribution obtained is used then as probability A priori in a traditional process of Inférence bayésienne