To appear: IEEE Transactions on Information Forensics and Security Face Recognition Performance: Role of Demographic Information

Subfields: Computer Vision and Pattern Recognition, AI Ethics, Fairness & Society

Citations in the corpus

• Model Cards for Model Reporting (2018, background): “ially problematic when models are used in applications that have serious impacts on people’s lives, such as in health care [14, 42, 44], employment [1, 13, 29], education [23, 45] and law enforcement [2, 7, 20, 34]. Researchers have discovered systematic biases in commercial machine learning models ”
• Designing Disaggregated Evaluations of AI Systems: Choices, Considerations, and Tradeoffs (2021, background): “the early 2000s, focusing primarily on performance for groups based on environmental factors like pose and lighting conditions (e.g., Lee, Ho, and Kriegman 2005; Beveridge et al. 2009; Klare et al. 2012) and for groups based on gender, age, and race (e.”
• Multiset-Equivariant Set Prediction with Approximate Implicit Differentiation (2021, background): “It is of course especially important to be careful when the system concerns humans, for example with biases relating to human ethnicity and gender in facial recognition systems (Klare et al., 2012; Robinson et al., 2020).”
• Uncovering and Mitigating Algorithmic Bias through Learned Latent Structure (2019, methodology): “Positive examples were taken from the CelebA dataset (Liu et al. 2015) and cropped to a square based on the annotated face bounding box.”
• Racial Faces in the Wild: Reducing Racial Bias by Information Maximization Adaptation Network (2018, background): “Several studies [49, 29, 23, 50, 36] have uncovered racial bias in non-deep 1 http://www.whdeng.cn/RFW/index.html face recognition algorithms.”
• Taxonomizing and Measuring Representational Harms: A Look at Image Tagging (2023, background): “well as different combinations thereof (Buolamwini and Gebru 2018; Khiyari and Wechsler 2016; Klare et al. 2012).”
• Deep Face Recognition: A Survey (2018, methodology): “in applications when directly adopting the pre-trained models. Another universal but serious data bias is uneven distributions of demographic cohorts (e.g., race/ethnicity, gender, age). According to [109], [64], [17], the female, Black, and younger cohorts are usually more difficult to recognize for”
• Algorithms, Platforms, and Ethnic Bias: An Integrative Essay Forthcoming in Phylon: The Clark Atlanta University Review of Race and Culture (2018, background): “For instance, facial recognition software often does not have a diverse set of training data, which results in poor performance and low accuracy for the underrepresented sample, which often comprises minorities (Klare, Burge, Klontz, Bruegge, & Jain, 2012).”
• Themis: automatically testing software for discrimination (2018, background): “Facial recognition systems often underperform on female and black faces [14].”
• Interpretable Machine Learning (2019, background): “For example, research shows that if a facial recognition algorithm is trained solely on the faces of White people, it performs more accurately for that group than others.(41) Algorithms can also reflect historic biases that exist in the data they are trained on.”
• 50 years of biometric research: Accomplishments, challenges, and opportunities (2016, unknown): “However, this is challenging in practice because it is very difficult to isolate the effect of aging phenomenon from other types of noise affecting a biometric measurement [10] [43] [49] .”
• Learning Race from Face: A Survey (2014, background): “In particular, experimental results in [28], [29], [30], [31], [68], [169], [175], [220], [221], [222] have shown that both commercial and nontrainable face recognition algorithms consistently tend to perform lower on the same cohorts (females, African”
• Predictive Inequity in Object Detection (2019, background): “Early warnings that facial recognition might have higher accuracy on white men showed that this problem might be somewhat mitigated by training systems separately for different demographic groups (Klare et al., 2012).”
• Demographic Effects in Facial Recognition and Their Dependence on Image Acquisition: An Evaluation of Eleven Commercial Systems (2019, result): “In our study as well as in [7], mated similarity scores were lower for those identifying as Black or African-American.”
• Image Counterfactual Sensitivity Analysis for Detecting Unintended Bias (2019, methodology): “For example, in 2012, [22] found that facial recognition models performed worse on female faces, black faces, and younger adults.”