One in five Americans will be diagnosed with skin cancer before they turn 70, a disease frequently seen by physicians at the Sylvester Comprehensive Cancer Center.
And people of all ages are enamored with taking selfies.
University of Miami graduate student Di Lun hopes to capitalize on both these trends to educate the public about the most common form of cancer diagnosed in America. As a recipient of one of the first two predoctoral U-LINK fellowship grants, she is working to create a selfie filter that will use augmented reality to show people the harmful effects of sun damage and skin cancer. The second recipient, Rahul Dass, is developing a facial analysis algorithm that would more accurately identify the race and ethnicity of arrestees to help prevent bias in the criminal justice system.
Vice Provost for Research John Bixby said these two students were chosen as the first U-LINK fellows because they are tackling important societal problems, and are working with collaborative faculty mentors from different subject areas. U-LINK, or UM’s Laboratory for Integrative Knowledge, is an initiative of the University’s Roadmap to Our New Century, to encourage interdisciplinary cooperation across UM.
“One goal of U-LINK is to cultivate the practice of interdisciplinary research across the university,” said Bixby, a co-lead on U-LINK with Associate Dean for Research Susan Morgan. “These students are now ambassadors for the idea of an interdisciplinary approach and we expect that these fellows will seed new interdisciplinary collaborations. Even the students who did not get funding have already opened conversations among faculty from different disciplines.”
Dass and Lun’s projects were chosen from a pool of 41 applicants by the U-LINK Action Team, which is made up of faculty and staff from across the university, Bixby said. This summer marks the first time that U-LINK is awarding the $40,000 fellowships. Graduate students can use the grants to complete their dissertations, and they can be renewed for up to three years, he added.
Specifically, Lun’s dissertation project will use augmented reality to create a cell phone application that will project the effects of sun damage on people’s upper bodies. It also will feature information about the dangers of skin cancer, and actions people can take to protect themselves. Through her project, Lun aims to show the public the perils of tanning, so people will protect themselves more often.
“People tan to look better, but it actually does more damage than you think,” said Lun, a doctoral student in health communication, who earned a master’s degree in public health at East Tennessee State University before coming to UM. “By doing this project, I hope it will serve an educational and persuasive purpose, to let people know the consequences of overexposure to the sun, or to ultraviolet rays through indoor tanning.”
Meanwhile Dass, a computer science doctoral student, is working to improve current facial recognition programs, many of which struggle to accurately identify a person’s race or ethnic background well, and therefore often misrepresent people. For example, in a popular video, MIT Media Lab researcher Joy Buolamwini, an African American woman, shares how a few facial recognition programs did not even detect her face, and those that did often decided she was a male. Dass said this often happens because facial recognition programs are built with a database of mostly white, male, celebrity faces, so they malfunction when exposed to a more diverse group of faces.
“I’m trying to test and develop a model that can classify race and ethnicity of people as accurately as possible,” Dass said. “If in 2019, we cannot accurately identify an African American woman, you can see what a difficult problem this is.”
Using a database of 200,000 mugshots from Miami-Dade County’s jail system, Dass is trying to craft a machine learning program that will correctly distinguish between white non-Hispanic, black non-Hispanic, white Hispanic and black Hispanic defendants. Since law enforcement agencies are increasingly relying on facial recognition algorithms to identify suspects, and they are often inaccurate, biases in the technology may open the door to mistaken classification and wrongful arrests, said Nick Petersen, assistant professor of sociology and one of Dass’ faculty advisors. This project can help shed light on how algorithmic biases can creep into the criminal justice process, along with strategies to minimize the errors, Petersen added.
“Machine learning and facial recognition software are being increasingly used by government agencies and private companies, so to understand how these work… is key to understanding how we can move forward because the reality is it’s probably not going anywhere soon,” said Petersen, who is also working on a faculty U-LINK team to examine bias in the criminal justice system. “There are these biases in the software [that exist now] and it’s problematic.”
