Can software tell surgeons the success of a medical procedure before it is complete? Can a boat read a river? Can a robot pick a berry?
Over the past five years, Lehigh’s Department of Computer Science and Engineering (CSE) has brought students from all over the U.S. to Lehigh to answer questions such as these.
With support from the National Science Foundation’s Research Experience for Undergraduates (REU), these students join Lehigh graduate mentors and CSE faculty team leaders for what amounts to a 10-week taste of a graduate life.
Building C, which serves as the home of the University’s new Interdisciplinary Research Institute for Data, Intelligent Systems, and Computation, as well as the innovative Mountaintop Summer Experience, serves as the base of operations for the REU program.
“REU provides for students who don’t have the opportunity to otherwise engage in intense research projects,” says Brian Davison, an associate professor and principal investigator for Lehigh’s Intelligent and Scalable Systems REU site, “and Mountaintop is a perfect context to introduce talented young engineers to Lehigh.”
Students took a deep dive into fundamental topics in both machine learning and scalable computer systems, such as new algorithms for machine learning, new approaches to privacy preservation and new techniques for increasing computer performance.
Matthew Camp, a rising senior from Georgia Gwinnett College, and Kenneth Gonzalez from the University of Puerto Rico at Arecibo, were part of the REU team that used artificial intelligence (AI) to help brain surgeons immediately “see” how successfully they had removed a tumor—before finalizing the procedure and resealing the braincase. The team’s software transforms an MRI into an image that can be interpreted in an instant, allowing medical practitioners to more confidently determine their next steps.
“I want to use computers to their fullest potential in improving people’s lives,” says Matthew.
The team used AI to help identify tumor outlines in medical imaging data obtained from Brigham and Women’s Hospital, essentially teaching a computer to identify tumor outlines in three dimensions. By the end of the project, the team had achieved 82 percent accuracy, and theorized that the technique could be improved, and eventually identify previously unseen tumors at least as well as human technicians. Matthew reports that the intensive research was a positive experience, and convinced him to focus on AI in graduate school.
Nicole Wang of the Georgia Institute of Technology was part of another team that helped to enable autonomous boats to monitor and survey the conditions of the Lehigh River. While others improved upon mechanical designs, added sensors, and developed navigation schemes, Nicole’s task was to incorporate a camera to detect and avoid objects. Her goal was to understand how the camera, capable of capturing images in 3D with an almost 270-degree view, worked, and how to create a map from that data to reveal obstacles.
“Two of my favorite research topics of computer science are robotics and human computer interaction,” says Nicole. “This research project topic was all about robotics, and I like the idea of developing technology that makes our lives easier.”
A third team, which included Nikolas Lamb from Clarkson University, helped to build a neural network to identify strawberries in a field. His team programmed a small, single-board computer to detect the berries, and then instruct a robotic arm to gently pluck the delicate fruit. The team programmed and refined the network to make it a faster detector without sacrificing precision, achieving 85 percent accuracy by training the machine with images of ripe red strawberries on a green leafy background.
“It’s one giant linear equation you have to adjust,” says Lamb. “The program can produce a prediction every 0.6 seconds, and we got to see what the machine was seeing.”
Nikolas believes that computer identification of objects is a relatively new but already well-explored field in AI, so there was a rich set of experiences to draw upon, and that his team’s research could be extended to other applications in agricultural production—different forms of produce, removing weeds from patches of more desired flora—and elsewhere.
His advisor, Professor Mooi Choo Chuah, encouraged him to apply to Lehigh for his Master's degree.
"I found REU students to be the type we want to recruit for our graduate program," she says. "Plus, undergraduate students blossom through exposure to project-based experiences; employers are aware of the value of this type of learning, as well."
“Before this program, I was on the fence about grad school,” says Chris Kjellqvist, a University of Rochester junior supervised by Associate Professor and REU program co-PI and Mike Spear, who teamed up with a Lehigh PhD student as well as researchers from Penn State University to explore scaling algorithms to handle ever growing mountains of data. His job was to “hack” a complex computer compiler to improve its ability to utilize advanced memory features.
“We worked to create new techniques that makes technologies such as parallel processing even stronger,” he says. “These are super-useful and innovative features, and we worked to provide programmers with ease of use.”
And perhaps just as importantly, Chris’ REU experience led to another significant outcome.
“After this project,” he says, “I know that this is the direction of computer science I want to pursue.”
