Mathematics of Machine Learning

This course presents the fundamental mathematical theory for modern machine learning techniques, including empirical risk minimization (error decomposition, Rademacher complexity, covering number), optimization method (gradient descent and stochastic gradient descent), kernel methods, and deep neural networks (approximation theory and over-parameterization). Students are expected to have knowledge in MATH2040/2048, MATH2050/2058 and MATH2060/2068, or equivalent.