Nigel Bosch - Teaching machines to teach humans: Data mining, emotion, and ethics

Abstract: In this talk, I examine the emotions people experience during learning and discuss methodologies for automatically predicting emotion and learning behaviors. Computerized learning environments, from online courses to smartphone applications, have become increasingly pervasive in modern educational contexts. These environments provide rich sources of data (e.g., videos, records of student actions) which can be mined to provide insights into the emotions and behaviors linked to effective learning. In turn, computational models derived from these data enable the development of interfaces that adapt to learners' needs. This talk focuses on the detection of emotion from facial features in videos, theoretical justifications for these methods, and how these methods can be improved in future work by leveraging recent advances in deep neural network architectures. Additionally, I discuss ongoing work which addresses ethical issues that arise when deploying machine learning models, such as how to identify and improve models that are more accurate for one racial group than another. In sum, this work shows how fair user models can be developed to understand and improve educational outcomes.

Nigel Bosch is a postdoctoral researcher at the National Center for Supercomputing Applications (NCSA) at Illinois. His research employs machine learning to model emotion, engagement, and learning, based on data from human-centered sources such as facial expressions and records of user's behaviors. This work has led to over 30 publications, 6 awards, and a wide variety of interdisciplinary collaborations. His current research interests are centered around egalitarian data mining methods (e.g., how can user models be made equally accurate for minority populations?), online education (e.g., do students' online interactions with each other predict success?), and related areas.

Questions? Contact Christine Hopper