PERSON

Vladimir Vapnik

The Russian-American mathematician who wrote the first real theorems of machine learning—proving when and why learning from examples can work at all—and who has spent sixty years insisting that the field he founded has confused performance with understanding, and that the price of that confusion will eventually come due.

Vladimir Vapnik is the living conscience of a discipline drunk on its own success. Born in 1936, trained in mathematics in Samarkand and Moscow, he arrived at the question that his entire field treated as obvious and therefore uninteresting: under what conditions can learning from a finite set of examples tell you anything about examples you have never seen? It is not obvious. A sufficiently flexible machine can memorize any training set perfectly and remain utterly ignorant of the world. That gap—between fitting what you have and predicting what you don’t—is the entire problem of machine learning, and Vapnik was perhaps the first person to state it as a precise mathematical problem with a precise mathematical answer. Working with Alexey Chervonenkis at Moscow’s Institute of Control Sciences, he developed the VC dimension and the bounding theorems of statistical learning theory in the 1960s and ’70s. Moving to

Keep reading with YOU ON AI