Machine Learning for Adaptive Learning
An effective adaptive learning system must identify deficiencies, provide appropriate remediation, and report actionable insights to students, teachers, and administrators. Machine learning aids in this mission by allowing us to infer aspects of students and educational content based on data, enabling any individual student on the system to benefit from the collective experience of many students. A critical task benefitting from this approach is estimating a student’s proficiency in one or more conceptual areas based on their and other students’ previous activity on similar content. In this talk, I’ll review this problem, its associated challenges, and some of the solutions that have been proposed. I’ll then present some recently published results comparing a class of Bayesian models based on item response theory with a deep learning approach employing recurrent neural networks.
Yan comes to Knewton with a strong background in computation, machine learning, and statistical data analysis. He got his Ph.D. in Computer Science at Carnegie Mellon University, where he was also involved in the Center for the Neural Basis of Cognition, an interdisciplinary program covering computational neuroscience and related fields. He received the prestigious Department of Energy (DOE) Computational Science Graduate Fellowship, and spent time at the DOE Lawrence Livermore National Lab developing sophisticated machine learning methods for automatically predicting RNA function. Prior to joining Knewton, he was a Howard Hughes Medical Institute fellow at New York University, where his research focused on understanding statistical patterns in images and sounds, and on formulating theories of information processing in the brain. In 2009, Yan published research in Nature and has taught courses in both computer science and neuroscience.