Prof. Marian Stewart Bartlett

Marian Stewart Bartlett, University of California, San Diego & Apple Inc.

Bio: Marian Bartlett was Full Research Professor at the University of California San Diego, and Co-Founder of Emotient, a start-up for automatic facial expression recognition. At UCSD, she co-directed the Machine Perception Lab at the Institute for Neural Computation, where she studied machine learning and computer vision, with application to face recognition and expression analysis. She received her undergraduate degree in Mathematics from Middlebury College and her Ph.D. in Cognitive Science and Psychology from UCSD. Dr. Bartlett authored over 80 articles in scientific journals and peer reviewed conference proceedings, and has 5 patents. Her book, Face Image Analysis by Unsupervised Learning, describes her work applying principles of information theory to face recognition with Terry Sejnowski at the Salk Institute. Dr. Bartlett collaborated with Paul Ekman and Javier Movellan to automate the Facial Action Coding System, and employed the system to study spontaneous facial expressions in areas such as learning and education. She served as Director of Scientific Programs for the Temporal Dynamics of Learning Center, a multi-institution NSF Science of Learning Center, as Associate Editor of the Neurocomputing journal, as Program Co-chair for IEEE Face and Gesture Recognition in 2011, and as General Co-Chair for the International Conference on Development and Learning in 2004. Dr. Bartlett was selected to the 2016 Wired Next List, and received the Women who mean Business Award from the San Diego Business Journal in 2014. She has served on the Board of the Foundation for Neural Information Processing Systems as Treasurer since 2007. Dr. Bartlett is presently a Research Scientist at Apple.

Title: Spontaneous Facial Expression Recognition: Insights for Reinforcement Learning

Abstract: Spontaneous facial expressions have the potential to measure latent variables in reinforcement learning. Facial expressions are the product of at least two distinct neural systems. Deliberate expressions originate in the primary motor cortex, whereas spontaneous expressions originate in the limbic motor area, a brain region that has been associated with fast value assessment, and follow a separate neural pathway to the face that is faster and more reflexive in nature. The muscles of facial expression originate from vestigial gills that are still present in the human embryo, and migrate over the face during development. Darwin theorized that facial expressions evolved from reflexive actions related to respiration, sensing, basic drives, and approach-avoidance responses, which were later shaped by natural selection to aid social communication. Capturing facial expressions from the spontaneous pathway has the potential to reveal rapid value assessment and affective substrates of learning. In this talk I will review considerations for developing a system to detect spontaneous facial expressions. I will then describe research using automatic detection of spontaneous expressions to measure latent variables for reinforcement learning. Facial responses were associated with reward-prediction-error and Shannon entropy. The facial results also support asymmetric reward learning algorithms, since facial response to positive and negative prediction errors are not symmetric opposites. Negative facial responses were associated with more volatile strategy switching. Scaling the learning rate by facial responses improved predictions by the reinforcement learning model. Facial expression measurement also provided insight into dysfunctions in learning. These results have implications for affectively aware automated tutoring systems that respond to the student’s emotional state the way good teachers do.

This talk reviews research conducted at University of California, San Diego, prior to joining Apple. It does not cover work conducted at Apple. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author and do not represent the views or opinions of Apple Inc.