Vladimir Vapnik

Co-developed Vapnik-Chervonenkis theory in the 1960s and 70s, the mathematical framework for how much a learning algorithm can be trusted to generalise beyond the examples it was trained on — still the theoretical backbone of statistical machine learning. Moved from Moscow's Institute of Control Sciences to AT&T Bell Labs in 1990, where he and Corinna Cortes turned that theory into the support-vector machine, one of the most widely used classification algorithms of the following two decades. Has taught at Columbia since 2003.