Data Modeling (Reinvented)

Previous ISyE Assistant Professor Ying Cui wants to help the world make better use of Big Data.

While stuck at home during the Covid-19 lockdown, Ying Cui saw a silver lining to her situation. Cui, who joined the ISyE Department as an Assistant Professor just weeks before lockdowns began, used the time to complete an 800-page book with her postdoc advisor, Jong-Shi Pang from the University of Southern California. Researchers in a wide range of disciplines have already taken notice as Cui’s book can be used to learn how to make accurate predictions from oceans of data.

“In order to solve an optimization problem, most likely you need numerical algorithms,” Cui says.

These algorithms, she explains, could and should be as adaptive as available data permit. Currently, researchers and analysts use simple linear models or heuristic algorithms to capture information from data. But, Cui says, “the world is not linear.” Which is why she intends to use her book and ongoing research to help others make more sophisticated use of “Big Data.”


‘The world is not linear’

It’s human nature, Cui acknowledges, to want to “simplify reality. But when we do that, we really lose some accuracy.” Her book provides mathematical foundations for nondifferentiable and nonconvex optimization problems. It embraces data with lots of variability, which contain sharp points or corners. She uses the letter “M” in the University of Minnesota’s logo as a metaphor for such models.

“You deal with some changing points; it goes up and down,” says Cui. “If you imagine data doing this same up-and-down, you have to imagine that you cannot manage it like a straight line. You have to take a different approach.” Using nonconvex and nondifferentiable methods to make meaning from raw data is daunting and complex stuff—computational ease is sacrificed for realism—and capable of yielding more accurate, precise predictions. And consequently, Cui says, “resulting in better policies and decisions.”

Much of Cui’s early career was focused on mathematics for mathematics’ sake. “When I was younger, I knew I liked to study mathematics,” Cui says. It wasn’t until her post-doc work began that “I realized all this mathematical stuff I worked on before could be useful in the real world.” Cui is excited to help people in a range of fields—from civil engineering to health care to economics—apply her work to advance their own research.

And last fall, the mathematical models Cui developed allowed one of her students to research optimal substance and timing of government responses to the pandemic using real data from the state of Minnesota. Cui’s book and her expertise helped Anthony Zhenhuan Zhang complete his final Ph.D. project—which yielded invaluable potential guidance for policymakers—while earning runner-up for best paper from the Production and Operations Management Society’s College of Humanitarian Operations and Crisis Management.

“The reason I sought Ying’s help with this paper is because I specifically wanted to leverage [her] optimization techniques to solving these problems,” Zhang says. Responding effectively to Covid-19, he adds, “is a research question that everyone cares about, and we were able to provide strong theory support with these techniques.”

“I have students from different fields and backgrounds, and some of them are talking about new applications for my research that I have never considered.”

—Ying Cui, ISyE Assistant Professor

Cui found the collaboration highly satisfying. “We tried to model the disease dynamic, and to understand how governments [might] affect the progress of the disease,” she says. They explored the cost effectiveness of various government policies—lockdowns, mask mandates, social distancing—considering factors like timing, unintended consequences, and how people might react to those policies.

The goal was to discern which actions—and at which times—might effectively slow disease spread with the least economic disruption. Zhang and Cui, with their fellow researchers, discovered that in Minnesota social distancing policies to be more effective than lockdowns. Additionally, they learned it is critical to implement social distancing policies at the post-pandemic-peak as more infectious variants become dominant.


Manifold applications

The fast-growing field of personalized medical treatment, known as “precision medicine,” is one of many areas that stand to benefit from Cui’s nonconvex and nondifferentiable data models. Data detailing a patient’s unique medical background could be modeled in new ways to create better outcomes.

But that’s not all. Housing costs and utilities pricing can also be more accurately predicted with Cui’s approach. She expects to see myriad other applications in numerous disciplines, many of which she’s discovered through connections with colleagues in other colleges and departments at the University of Minnesota. “We can learn from each other and benefit each other’s work,” Cui says.

ISyE Professor Zhaosong Lu has known Cui for a few years. He was one of the reviewers for her dissertation and is also on Cui’s mentoring committee in the ISyE Department.

Lu imagines Cui’s work will have enormous implications in a wide array of fields. “This kind of research crosses different [disciplinary boundaries]—computer science, electrical engineering, machine learning and artificial intelligence, and statistical analysis,” Lu says. “She’s a rising star, and she has all these great characteristics to make her successful here.”

Last spring, Cui offered a Ph.D. course based on her new book and asked students to work on projects relevant to nonconvex optimization. She found their ideas exhilarating. “I have students from different fields and backgrounds, and some of them are talking about new applications that I have never considered,” Cui says. “Talking with them, understanding the problems [they want to solve], feeling their energy around all these things—it’s really rewarding!

“It motivates me and helps me to understand what is important in the applied area. I want to understand more applications, that way I know what else has not been developed mathematically and I can work on that next.”

Story by: Susan Maas