Ray Solomonoff

Ray Solomonoff

An American mathematician who attended the 1956 Dartmouth workshop and went on to invent algorithmic probability, assigning to any observed sequence a prior probability based on the length of the shortest program that could produce it. Combined with Bayes' rule, this yields a formal theory of universal inductive inference: the best prediction is the simplest explanation consistent with the data. Developed alongside Kolmogorov and Chaitin's work on algorithmic information theory, it is the sharpest mathematical statement of Occam's razor and an idealised limit for what any learning system could achieve. (See also: Thomas Bayes, Machine learning, Overfitting)