Riya Tyagi, a 2023 NASA SEES intern on the Earth Systems Explorer team, was selected as a finalist in the 2024 Regeneron Science Talent Search. The finalists represent 36 schools across 19 states and are competing for more than $1.8, with a top prize of $250,000.
Riya’s project title is: “Using Computer Vision To Disentangle Features Enabling AI To Learn Self-Reported Race and Ethnicity From Medical Images”
Riya Tyagi, 16, of Exeter, sought to understand how convolutional neural networks have learned how to determine patients’ self-reported race and ethnicity for her Regeneron Science Talent Search computer science project. It currently remains unclear how machine learning, when analyzing medical image data, can determine a patient’s race and ethnicity. Understanding this is key to minimizing bias when A.I. tools are used for medical purposes. By training hundreds of convolutional neural networks, a specific neural network that quickly and precisely analyzes images, Riya investigated what kinds of information the networks were using to determine patients’ race and ethnicity. After the system reviewed 11,000 images of the human retina, Riya determined which key features were responsible. By analyzing these features, she discovered a potential cause of bias: the U-Net segmentation model widely used for medical image processing. She hopes her work will allow development of ethical A.I.-powered healthcare software to minimize racial bias in the future. The daughter of Urvashi and Shishir Tyagi, Riya attends Phillips Exeter Academy, is in Science Olympiad, and co-led her varsity robotics team to a second-place win out of 6,400 teams at a World Championship. She also founded one of the first all-girl Boy Scouts troops and was inducted into the Order of the Arrow.
Source: https://www.societyforscience.org/regeneron-sts/2024-finalists