Mike Wu

Recent revolutions in artificial intelligence have led to a series of promising findings in computer vision, natural language, and more. However, many of the popular algorithms demand millions of annotated examples, an expensive bottleneck that is often unrealistic in many application domains, such as education. Students tackle the same problem in thousands of different ways and even the biggest classrooms only accommodate a thousand students, making it impossible to collect the amount of annotated data that modern algorithms require. The goal of my research is to develop new "un-supervised" algorithms, that can more efficiently model small datasets without annotation. With this, I look to apply these new algorithms to understanding student learning from a computational lens. In particular, I am exploring methods to measure student ability, and to automate feedback for student work—two important ingredients to bring high quality education to scale.