Colloquium: Automated Interventions to Improve Student Learning: Reinforcement Learning Based Approaches
October 21, 2019 - 11:15am to 12:15pm
3-180 Keller Hall
Online educational technologies offer learners the opportunity to receive interactive feedback, with the potential to address misunderstandings or re-explain complex concepts. This feedback can be personalized based on this learner's or previous learners' interactions with the technology. In this talk, I’ll describe two projects that use algorithms related to reinforcement learning to provide more effective feedback. In one project, we developed a variation of inverse reinforcement learning to examine learners’ behaviors and make fine-grained assessments about their (mis-)understanding; choosing interventions based on inferences about learners’ specific misunderstandings improves their later performance. In the other project, we took a model-free approach in which we treated the choice of intervention in a MOOC as a multi-armed bandit problem, and explore the impact on learners as well as on researchers studying what intervention are most effective. Both of these projects point to the potential of computational methods for creating smarter educational resources that address individual students' needs, and in each project, consideration of the specific educational problem has led to investigation of computational questions that may be more broadly applicable.
Anna Rafferty is an Assistant Professor of Computer Science at Carleton College. She earned her Ph.D. degree in computer science from UC Berkeley. Her research focuses on combining computational cognitive science and machine learning in educational applications. Specifically, her work has emphasized characterizing problems faced in educational technologies using more general computational frameworks, and has also examined how computational methods can be used in technologies in the classroom, such as work exploring the use of automated guidance about students' scientific drawings.