When he was younger, Cyrus Lee never thought he would pursue a career in STEM, but that mindset changed during his time in high school.
Looking to challenge himself academically, Lee sought out the most difficult courses available. He gravitated toward mathematics initially, and soon began to branch out to related fields, like engineering and programming.
His interest in STEM followed him to Vanderbilt University, where he is now a double major in computer engineering and math.
Eager to learn more about deep learning, a field of artificial intelligence, Lee spent 10 weeks recently at Oak Ridge National Laboratory (ORNL) as a participant in the Higher Education Research Experiences (HERE) program.
Under the mentorship of Travis Johnston, Ph.D., a staff researcher on the Nature Inspired Machine Learning team, Lee used deep learning techniques to map out objects in commercial satellite images as part of Project SOFIA (Scarce Object FInding Algorithm).
“The general idea behind Project SOFIA is to define an object to be searched for (like solar panels or ice cream trucks) by providing a single example to a machine learner and finding all instances of the object within a particular image or set of images,” Lee explained.
In the field of deep learning, neural networks are given the ability to “learn” how to distinguish features and patterns in vast datasets. In the case of Project SOFIA, those datasets are commercial satellite images. At the conclusion of his internship, Lee demonstrated that neural networks trained on consumer imagery (like that from cellphones) were not reliable object detectors when deployed on commercial satellite imagery. The strategy of training neural networks on one type of data (consumer imagery, for example) and deploying the network on different data is a commonly accepted and utilized approach in industry. However, to recognize objects in the commercial satellite pictures more consistently, researchers will need to create a network trained specifically on commercial satellite imagery.
The research tool used in Project SOFIA has many potential real-world applications, Lee noted.
“In our ever-changing world, with novel data springing from all parts of the globe, having a mechanism with which to locate a specific object or feature across commercial satellite images becomes incredibly useful,” Lee said. “Given the plethora of images amassed by commercial satellites, finding every occurrence of a specific object across these images would be nearly impossible without this tool. With this tool, geographers could locate building footprints, scientists could identify trends across the earth by counting objects and noting their locations, and the common user could locate notable geographic features.”
During his internship, Lee met regularly with Johnston to discuss Project SOFIA, with Johnston offering advice and feedback. Additionally, Lee often attended seminars where other researchers would share their findings, exposing him to new ideas on research and technology.
When Lee wasn’t delving into his research at the lab, he was exploring East Tennessee.
“Coming from the urban sprawl of Houston, I was excited to live in the small town of Oak Ridge for the summer,” he said. “I went hiking around the area when I was younger and yearned to return to the beauty and tranquility of the wooded mountains once again. Spending every day around the magnificent forest is a blessing that has helped me clear my head from the school year and focus on my research and beyond.”
Lee appreciates the opportunity to conduct research at ORNL through the HERE program, and he recommends the program to other students looking to gain real-world experience in their chosen field.
“This experience has benefited me by furthering my understanding of deep learning, as I have had the opportunity to learn more about neural networks, analyzing and modifying them through trial and error. I feel prepared to meet any deep learning challenge head-on,” he said.
The HERE program at ORNL is administered by the Oak Ridge Institute for Science and Education (ORISE) for the U.S. Department of Energy.