All attending in person are required to be masked per the current University policy. Zoom link will be available for those interested in attending the seminar virtually.
Intelligent human-centered learning seeks to model how people think, feel, and behave in various contexts. Enabling computers to better understand people in everyday situations has immense prosocial potential, but recent efforts are finding that these gains may only be realized by studying and learning from people in natural settings, outside of controlled laboratory experiments.
In this talk, I will highlight primary challenges facing human studies in naturalistic scenarios and will explore avenues for improving the validity and generalizability of AI assessments of human constructs.
In particular, this talk will examine case studies in the domains of health care, education, and justice, and I will present methods for improving data quality, experience label quality, and the applicability of AI assessments to different groups of people.
We will conclude by considering some exciting applications for human-centered learning once sufficient validity and generalizability have been achieved.
Brandon Booth is a postdoctoral research associate at CU Boulder working in the Emotive Computing Lab. His research focuses on using multi-modal machine learning techniques to model human perception, behavior, and experiences and developing algorithms to reduce the impact of inadvertent human biases and errors. He recently received his Ph.D. in computer science from the University of Southern California, and his work on continuous annotation fusion won the ACM AVEC gold-standard emotion representation challenge. He has a diverse industry background developing video games, serious games, robots, computer vision, and human-computer interaction systems.
To attend virtually, email lucyinstitute@nd.edu to request the zoom link.