There is a near relationship concerning machine learning and compression. A system that predicts the posterior probabilities of the sequence presented its complete historical past can be employed for optimal data compression (by making use of arithmetic coding to the output distribution).