Abstract
Thelastfive decades have seen the construction of a new branch of statistical physics that studies highly disordered systems. Starting with the study of spin glasses, this field has expanded to cover complex systems in various branches of science, from computer science to biology and information theory.Four main obstacles had to be overcome to develop the theory of disordered systems in very high dimensions : studying statistical sets of samples, quantitatively analyzing microscopic disorder, exploring complex energy landscapes, understanding their links with dynamical properties. This talk begins with an overview of these developments. It thendescribes the new challenge posed by the application of these methods in machine learning, that of structured disorder.