Reducing bias through the careful curation of training data for developing face recognition algorithms
Open Access DepositedAbstract of Praxis Reducing bias through the careful curation of training data for developing face recognition algorithms While face recognition algorithms have their uses, they also carry the potential to be biased and do harm to some groups. In order to create better models, it is essential to comprehend how these algorithms are trained and what factors affect their accuracy and fairness. How the ethnic composition of training data affects the performance of facial recognition systems is the topic of this research. We show that gender mislabeling and racial bias are two of several biases that can have a major effect on accuracy when combined. As part of our research, we compiled information from a wide variety of sources. We were able to significantly lower the likelihood of bias due to certain groups' underrepresentation, particularly African Americans, by including a wide range of perspectives and experiences in the data. We find that changing a color image to a grayscale one yields different meanings, which allows us to assess how well our curated datasets analyze the performance of our FRT algorithms. In order to implement our findings, we developed a Predictive Code Completion Model using Python code that made use of the Tkinter toolkit.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.