One Label, One Billion Faces - Usage and Consistency of Racial Categories in Computer Vision
Association for Computing Machinery (ACM) via YouTube
Overview
Explore a critical analysis of racial categorization in computer vision systems through this 15-minute conference talk presented at FAccT 2021. Delve into the complexities of face recognition technology, examining issues of group fairness, demographic parity, and the problematic nature of racial categories. Investigate the challenges of cross-dataset generalization and the perpetuation of stereotypes in AI systems. Gain insights into the ethical implications and limitations of current approaches to racial classification in machine learning, and consider potential solutions for improving fairness and accuracy in computer vision applications.
Syllabus
Intro
Face Recognition
Synthesis
Group Fairness
Demographic Parity
Fairness is based on groups.
Racial Categories: Badly Defined
Moment of Identification
Scenario 2
Classifier Ensemble
Cross-Dataset Generalization
Stereotypes
Conclusions
Taught by
ACM FAccT Conference