| Apr 07, 2026 | The research of Niklas and me investigated why AI models for facial expression recognition exhibit inconsistent performance across different subject populations, including healthy individuals and patients with facial palsy. Testing state-of-the-art models, the authors found that factors such as sex, age, and facial symmetry significantly skew model predictions, for instance, “happy” was more strongly activated for women, while “disgust” was systematically underweighted for men. The findings are featured in a press release (German, English) on the official webpage of the Friedrich Schiller University Jena. They were originally published in the paper “The Power of Properties: Uncovering the Influential Factors in Emotion Classification” at ICPRAI 2024. |