Data Bits & Bites - James Scott, SDS Chair

James Scott is Professor and Chair of the Department of Statistics and Data Sciences at The University of Texas at Austin, where he has been on the faculty since 2009. His research focuses on the theoretical and methodological foundations of statistics and machine learning. He has also conducted interdisciplinary research in health and public policy, including pioneering work in women’s health and in real-time disease tracking and forecasting during the COVID-19 pandemic. Scott is the co-author of AIQ: How People and Machines Are Smarter Together, which The Wall Street Journal described as “a book about AI with a moral core.” He studied mathematics at University of Cambridge as a Marshall Scholar and earned his Ph.D. in Statistics from Duke University. His honors include the NSF CAREER Award, the Susie Bayarri Award for outstanding young researchers in Bayesian statistics, and the UT System Regents’ Outstanding Teaching Award. 

Title: Vector Factories and Hyperplanes: How Geometry Powers Machine Learning

Abstract: How can a computer learn to reason about language, recognize images, or make predictions from messy real-world information? Before any learning can begin, a more basic problem must be solved: how can words, pictures, and other awkward forms of data be represented in a form that mathematics can work with?

In this talk, I explore two key geometric ideas behind that translation. The first is what we might call “vector factories,” which in machine learning are called “embeddings": methods that convert raw, unstructured inputs into vectors of numbers that can be analyzed mathematically. The second idea is hyperplanes: once data are represented as points in a high-dimensional Euclidean space, learning often amounts to drawing simple boundaries that separate different kinds of points and allow machines to classify, predict, and decide.

Machine learning is far broader than these two ideas, but a remarkable amount of it can be understood through this geometric lens. This talk offers an intuitive and visual introduction to how modern machine-learning systems are built from these simple ingredients.