Yi Han, speaking at the Institute for Advanced Study's Matrix Pizza Night seminar, presents work on products of random matrices and their scaling limits. The talk traces how multiplying many random matrices together produces predictable statistical behavior as the number of factors grows, connecting classical results in random matrix theory to the layered structure of deep neural networks, where each layer's weights act like another random matrix applied to the data. Han walks through the mathematical machinery used to control these products, including limiting spectral distributions and Lyapunov exponents, and discusses what the resulting scaling limits suggest about signal propagation and stability in deep networks. Held in Rubenstein Commons Room 5, the session runs nearly ninety minutes and is pitched at an audience with working knowledge of probability and linear algebra, moving from rigorous setup to applications in modern machine learning theory.