Modern and Emerging Phenomena in Machine Learning

Abstract

Modern machine learning systems exhibit a range of phenomena that are not adequately explained by classical theories. The workshop examined recent progress toward a mathematical understanding of these phenomena, bringing together researchers with expertise in probability, analysis, geometry, optimization, statistics, and theoretical computer science. Discussions addressed complexity and parametrization, optimization dynamics, representation learning, robustness, and the mathematical structure of modern architectures. Particular emphasis was placed on identifying intrinsic geometric, probabilistic, and computational principles underlying learning systems. The workshop fostered interactions across disciplines and highlighted a number of open problems and future directions for the mathematical foundations of machine learning.