"La théorie de l'apprentissage de Vapnik et les progrès récents de l'IA" par Yann Le Cun

The recent explosion of artificial intelligence and its applications is largely due to the use of machine learning, particularly deep learning techniques. The foundational ideas of machine learning date back to the 1950s, and those of deep learning to the 1980s. Practical applications have exploded in recent years, from self-driving cars to translation, information retrieval, and personalized medicine. But what does the theory say? The most general mathematical formulation of learning processes is that of the Russian mathematician Vladimir Vapnik, who emigrated to the United States. In his short book published in 1995, "The Nature of the Statistical Theory of Learning," Vapnik describes the general conditions under which a system, computer or biological, can learn a task, and at what speed. But this general theory does not explain why modern deep learning methods work so well, far better than simpler but mathematically better understood methods. Lecture in the series "One Text, One Mathematician." April 4, 2018 at the National Library of France.