machine learning is something like a process of probability, where a machine build up experience from previous results of the same process and then react.
Much recent work in machine learning has focused on learning good feature representations from unlabeled input data for higher-level tasks such as classification.