MACHINE LEARNING THEORY AND STATISTICAL MECHANICS

Course Description

This is a colloquium on the influence of physics on both foundational and cutting edge machine learning methods. We will explore how statistical physics can be used to build intuition for and prove properties of machine learning models and algorithms. We will examine the most foundational neural networks, like the multilayer perceptron, and the state-of-the-art generative models, like energy based models and diffusion models.

This colloquium is intended for Computer Science, Physics and Math students. Familiarity with calculus, linear algebra, and probability will be helpful, as will basic conceptual understanding of machine learning, but we will go over the relevant ideas.

This is a student-led colloquium led by Sidharth Kannan under the supervision of Eric Vigoda.