David Chapman Has a Passion for Image-Related Research

David R. Chapman, University of Miami Computer Science Professor

David Chapman Has a Passion for Image-Related Research

As a child, David Chapman spent hours in front of his family computer, playing StarCraft, Spyro the Dragon, and other video games that used high-quality graphics to take players on fantastic adventures. Chapman was intrigued, but not just with the games. He was captivated by the imagery, and wanted to understand how computer-generated images were made.

At just 10 years old, Chapman was already skilled in math, but he was soon inspired to make his own video games. He taught himself computer programming, and just two years later—at age 12—he started college with plans to major in computer science. His fascination with the importance of images—and with crafting computer algorithms that use them to answer questions in medicine, meteorology, and engineering—has permeated Chapman’s career. In fact, his training in computer vision techniques helped Chapman create remote sensing software to detect leaks in oil pipelines and computer programs that can help doctors identify conditions like tuberculosis from X-rays.

Oceaneering pipeline inspection robot

“Images are incredibly important to graphics, satellites, underwater vehicles, and health care, to name a few areas,” Chapman said. “So, while I’ve worked on many different applications, I’ve always been interested in image analysis algorithms.”

Yelena Yesha, speaker, Meet A Data Scientist Lecture Series, University of Miami Institute for Data Science and ComputingJust a year ago, Chapman joined the computer science faculty at the University of Miami, in the College of Arts and Sciences. Since then, he has collaborated with colleagues across the University to help accelerate research projects. Chapman also enjoys teaching students how to formulate their own computer algorithms to delve into data and uncover new insights.

This summer, Chapman was named the first Knight Foundation Junior Chair in Data Science and Artificial Intelligence at the Frost Institute for Data Science and Computing (IDSC). It is the second of six endowed chair positions the John S. and James L. Knight Foundation funded to foster more data science and artificial intelligence research at the University. In the role, Chapman will be working closely with his former mentor at University of Maryland, Baltimore County (UMBC), computer science professor Yelena Yesha, who holds the only other named Knight Chair at IDSC.

“David is an asset to IDSC, and also the whole University,” said Nick Tsinoremas, director of IDSC, vice provost of research computing and data, and a professor of biochemistry and molecular biology, focused on bioinformatics. “He can collaborate with climate and physical scientists just as well as those in the biomedical field. But he also has expertise in image recognition and computer vision, so he is knowledgeable about very cutting-edge technology and is a young and promising scholar in that field.”

Nick TsinoremasAfter graduating from college at 17, Chapman stayed on to finish his master’s degree at UMBC. For his thesis, Chapman used image processing methods from computer graphics to improve the resolution of infrared satellite images. “It was the first time I had used what I learned about video games to do something related to scientific computing,” he said.

Two years later, with his doctorate complete, Chapman moved to New York to do a postdoctoral fellowship at Columbia University’s Lamont Doherty Earth Observatory. As part of a group working on forecasts about the El Niño Southern Oscillation Cycle, an atmospheric phenomenon that drives global climate, Chapman helped create algorithms that can predict how the ocean temperatures could change six to eight months ahead.

He went on to land a job at Oceaneering International, where Chapman helped the company develop software for subsea robots that monitor oil pipelines in the Gulf of Mexico and ensure they are never leaking. If the pipelines do have any damage, the problem is detected and repaired. At Oceaneering, Chapman also gained a passion for teaching. During that time, he started his own summer camp in collaboration with the Gleger Center of Mathematics—which he attended as a child—to teach young scholars how to code and develop their own video game graphics like he did. Many former campers went on to major in computer science, Chapman said.

Before joining the University of Miami, Chapman worked as an assistant professor at UMBC, where he continued to capture the attention of his students, said Yesha. Much of his research focused on improving medical imaging analysis, and Chapman created programs to help radiologists diagnose lung illnesses, including two of the world’s most fatal infectious diseases—tuberculosis and COVID-19. “He is a very effective educator, and students enjoy working with him,” Yesha said. “He also has great mathematical underpinnings that make him well-equipped to attack difficult problems with novel algorithms.”

Computer Vision

Computer Vision + Neural Networks

Today, Chapman spends most of his time focused on an area of artificial intelligence called computer vision, which helps computers find meaningful information from images. It also includes image recognition, or the ability for computers to sort through pictures and quickly find specific objects, people, or other entities. There are a variety of ways to accomplish computer vision tasks, but Chapman said most scholars and private companies today utilize something called “neural networks.” These are massive collections of images or data that help train computer algorithms to do something. For example, if you want to teach the computer to find a cat in a photo, the neural network would likely have millions of animal photos, some with cats, and some with other animals, so the computer can learn the difference.

While neural networks is not a new idea in computer science, Chapman said that the explosion of data in recent years has helped make artificial intelligence tools—like machine learning—more powerful and efficient. Neural networks are what power things like ChatGPT and other large language models, he said. “Machine learning is about using data to help machines do things that are considered to be intelligent,” Chapman said. “Neural networks are one approach to machine learning, but they are now dominating other approaches because of their accuracy and because we have so much data today.”

It’s important to find ways of improving data interpretation
so that these tools are no longer overconfident
when the data is outside their realm of understanding.

But Chapman wants to improve upon these tools. In working on computer vision algorithms, he has noticed they often spit out a definitive answer, even if the new dataset is different than the one they were trained on. Often, this means the results are biased. For example, the race, background, or ethnicity of people in a health care dataset could be different than the one it was trained on, causing a skewed result. But bias could also mean the equipment that collected the data, say an X-ray machine from a different company, or data collected with different parameters could mean the new dataset fed into the program is very different than the one it was trained with. Any differences in the way that data is collected as opposed to how it is interpreted can introduce an algorithmic bias, Chapman said. Therefore, Chapman wants to find a way for these algorithms to recognize they are working with a very different dataset and pause before immediately offering an inaccurate result. “It’s important to find ways of improving this so that these tools are no longer overconfident when the data is outside their realm of understanding,” he said.

Yet, Chapman is also using his computer vision expertise to help colleagues in the College of Engineering create a program that will detect microcracks in concrete. Finding these early will ideally help buildings maintain their structural integrity before they become larger issues, he said. “We are using CT scans to do this, which is an interesting application of the concept,” he added. And while Chapman is still building his career, he credits his parents’ encouragement for his success. When he expressed an early interest in math, they sent him to the Gleger Center for extra tutoring. As a result, he was able to take and pass college entrance exams at the age of 11. Having that strong foundation in math helped him translate those skills to computer science, he said. “Being able to take a word problem, and break it into something mathematical that a computer can understand is a big part of what I do,” he said. “Getting that exposure to math at a younger age helped me with everything related to computer science.”


SOURCE:  NEWS@TheU Story by Janette Neuwahl Tannen