Abstract
Despite the rapid evolution in the development and deployment of machine learning methods in almost all areas of society, our theoretical understanding of the mechanisms behind this success remains rather limited. As discussed several times in this cycle, one reason behind this failure is the incompatibility of classical statistical methods in high dimensionality, also known as " the curse of high dimensionality ". Curiously, this is precisely the regime studied by statistical physics since the late eighteenth century in the context of many-body problems. The very natural connection with learning, due to the pioneering work of Elisabeth Gardner, Bernard Derrida and other physicists in the 1980s, is the basis of a rich symbiosis lasting right up to the present day.
