Vladimir Vapnik is a Soviet-born mathematician and computer scientist best known for co-founding statistical learning theory and for co-inventing the support vector machine (SVM), one of the most widely used algorithms in Machine learning before the resurgence of Deep learning after 2012.
Statistical learning theory
Born in 1936 in the Soviet Union, Vapnik worked at the Institute of Control Sciences in Moscow, where beginning in the 1960s he and Alexey Chervonenkis developed what became known as VC theory, after their initials. VC theory provides a mathematical framework for understanding when a learning algorithm trained on finite data will generalize well to new, unseen examples, formalized through the concept of VC dimension, a measure of how complex a class of functions is. This work gave the field of Supervised learning a principled theoretical foundation for reasoning about the tradeoff between fitting the training data closely and avoiding Overfitting, a concern later addressed in neural networks through techniques such as Regularization and dropout.
Support vector machines
After moving to the United States in 1990 and joining AT&T Bell Labs, Vapnik and colleagues including Corinna Cortes developed the modern support vector machine in the early 1990s, extending earlier linear classifier ideas with the "kernel trick," which lets an SVM find nonlinear decision boundaries by implicitly mapping data into a higher-dimensional space. Backed by VC theory's generalization guarantees, SVMs became a dominant, mathematically well-understood alternative to neural networks through the 1990s and 2000s, used widely in text classification, bioinformatics and computer vision, during a period when Neural network research was out of fashion following the reduced funding of the second AI winter.
Later career
Vapnik later worked at NEC Laboratories, Facebook AI Research (part of Meta AI), and as a professor at Columbia University and University of London. His 1995 book The Nature of Statistical Learning Theory remains a standard reference. Although SVMs were eventually overtaken by deep neural networks for most large-scale perception tasks once sufficient data and compute made end-to-end Gradient descent training practical, Vapnik's theoretical framework continues to inform how researchers think about generalization, sample complexity and the limits of what a given amount of Training data can support.