AI Seminar 142: James M. Rehg
From Prasad Tadepalli
views
From Prasad Tadepalli
Abstract: Beginning in infancy, individuals acquire the social and communication skills that are vital for a healthy and productive life. Children with autism face great challenges in acquiring these skills, resulting in substantial lifetime risks. The diagnosis, treatment, and study of autism depends fundamentally on the analysis of child behavior. Standard methods for behavioral observation and coding are the backbone of research studies but are inherently coarse-grained and not easily scalable. In this talk I will present our research in developing AI models and computer vision technology to automate the measurement of social behavior from video. Our goal is to unlock the rich behavioral information that is present in video and make it available for large-scale data-driven modeling, phenotyping, and assessment. I will present several recent findings which demonstrate the feasibility of this approach. The first is a method for detecting eye contact during naturalistic face-to-face interactions, which has been shown to achieve human-level accuracy, and it's extension to more general social gaze behaviors. I will also describe progress in developing longitudinal models of language development in ASD with the goal of predicting the risk for minimally verbal outcomes. This is joint work with Drs. Nancy Brady, Cathy Lord, Rebecca Jones, Sophy Kim, Jenna McDaniel, Agata Rozga, Sangmin Lee, Junho Kim, Eunji Chong, and Fiona Ryan, and Ph.D. students Xu Cao, Vipin Gunda, Frank Yang, Max Xu, and Aditya Radhakrishnan.