Using Machine Learning to Disentangle Counter-Intuitive and Confounded Emotion Stereotypes from Neutral Faces

We trained three machine learning models to predict facial expression resemblance using low-level visual characteristics of the face. We then used these models to examine their predictive validity on human impressions and real-world behaviors. Results revealed both counter-stereotypic (race) and stereotypic (gender) patterns that provide insight into human judgments.