How can a machine learn from experience? Probabilisticmodelling provides a framework for understanding what learning is, and hastherefore emerged as one of the principal theoretical and practical approachesfor designing machines that learn from data acquired through experience. Theprobabilistic framework, which describes how to represent and manipulateuncertainty about models and predictions, has a central role in scientific dataanalysis, machine learning, robotics, cognitive science and artificialintelligence. This Review provides an introduction to this framework, anddiscusses some of the state-of-the-art advances in the field, namely,probabilistic programming, Bayesian optimization, data compression andautomatic model discovery.
Probabilistic machine learning and artificial intelligence(2015).pdf