Off-Campus Hollins Users:
To access this document, please click here to log in to our proxy server with your campus network user name/password (the same one you use to log into the campus network and your e-mail).
Loading...
Event Type
Research Presentation
Academic Department
Mathematics and Statistics
Start Date
5-4-2021 12:00 AM
Description
Facial recognition has been a breakthrough in the development of Neural Networks and Artificial Intelligence. However, when used in a real-world setting, rather than just a testing dataset, specific programs will misidentify women and people of color far more often than white men. As facial recognition becomes more widely deployed these mistakes can have serious consequences. When police departments use biased technology to find suspects, it can lead to wrongful arrests and even convictions, as in the case of Robert Julian-Borchak Williams. I plan to create a facial recognition program of my own in which I train the algorithm on a dataset that proportionally represents both men and women and people of color. To do this, I will be using Google Colab and TensorFlow to create Neural Networks that we can train on our given dataset. I suspect that with a proportionally represented dataset, the program will be more accurate than one trained on a dataset that is disproportionately white men.
Unfairness and Discrimination in Machine Learning
Facial recognition has been a breakthrough in the development of Neural Networks and Artificial Intelligence. However, when used in a real-world setting, rather than just a testing dataset, specific programs will misidentify women and people of color far more often than white men. As facial recognition becomes more widely deployed these mistakes can have serious consequences. When police departments use biased technology to find suspects, it can lead to wrongful arrests and even convictions, as in the case of Robert Julian-Borchak Williams. I plan to create a facial recognition program of my own in which I train the algorithm on a dataset that proportionally represents both men and women and people of color. To do this, I will be using Google Colab and TensorFlow to create Neural Networks that we can train on our given dataset. I suspect that with a proportionally represented dataset, the program will be more accurate than one trained on a dataset that is disproportionately white men.
Comments
Under the direction of Dr. Giancarlo Schrementi.