Blogs
Abstract
This paper investigates the evolution of demographic accuracy in the Lumen facial recognition Rank-one program from 2019 to 2023, focusing on age, sex, and race-based disparities. Utilizing data from National Institute of Standards and Technology (NIST) evaluations, the study employs a meta-analysis approach, analyzing false match rates (FMRs) across program versions 7, 11, and 13. The methodology involves translating color-coded FMR data into numerical values and subsequently analyzing equity measures to assess accuracy improvements over time. The data demonstrates notable trends: improvements in accuracy from version 7 to 11, followed by a decline in version 13. Results also show demographic biases with individuals of color, older age groups, and females, leading to heightened risk of misidentification. Overall, the study highlights the importance of addressing demographic disparities in facial recognition technology to promote equality in its implementation.
​
​
Academic paper: https://docs.google.com/document/d/12OZpJhyK916hJTw5EZITxIZwuDB9F6iv9nm6mDBJXJQ/edit?usp=sharing
Final presentation: https://docs.google.com/presentation/d/1-ZnIMfawsPnDnXnKgVQw8pkv9ZZ5-0lhjO-qM7QNLkI/edit?usp=sharing