MITIGATING BIAS IN FACIAL ANALYSIS SYSTEMS BY INCORPORATING LABEL DIVERSITY
DEFESA DE DISSERTAÇÃO DE MESTRADO – Programa de Pós-Graduação em Ciência da Computação
MITIGATING BIAS IN FACIAL ANALYSIS SYSTEMS BY INCORPORATING LABEL DIVERSITY
ALUNA: Camila Kolling dos Reis
ORIENTADORA: Dra. Soraia Raupp Musse
COORIENTADOR: Dr. Adriano Alonso Veloso
BANCA EXAMINADORA: Dr. Virgílio Augusto Fernandes Almeida (DCC/UFMG), Dr.
Rafael Heitor Bordini (PPGCC/PUCRS)
DATA: 04 de julho de 2022
LOCAL: Videoconferência
HORÁRIO: 15:00
Link para acessar a videoconferência
RESUMO:
Facial analysis models are increasingly applied in real-world applications that have significant impact on peoples´ lives. However, as previously shown, models that automatically classify facial attributes might exhibit algorithmic discrimination behavior with respect to protected groups, potentially posing negative impacts on individuals and society. It is therefore critical to develop techniques that can mitigate unintended biases in facial classifiers. Hence, in this work, we introduce a novel learning method that combines both subjective human-based labels and objective annotations based on mathematical definitions of facial traits. Specifically, our proposed method first generates new objective annotations, each capturing a different mathematical perspective of the analyzed facial traits. We then use an ensemble learning method, which combines individual models trained on different types of annotations. We provide an in-depth analysis of the annotation procedure as well as the datasets distribution. Moreover, we empirically demonstrate that, by incorporating label diversity to the decision-making process, our method successfully mitigates unintended biases, while maintaining significant accuracy on the downstream tasks.
