As a native of China, Minxing Zhang’s familiarity with the Atlanta region barely extends beyond Emory University’s campus.
But working with four other Emory College of Arts and Sciences undergraduates, he “explored” the region this summer through neighborhood demographic data ranging from poverty level to race. As part of the experiential learning DataThink project in the Department of Quantitative Theory and Methods (QTM), the students used the data to build a model that quantifies the metro Atlanta rental landscape.
They recently presented their findings — showing rental prices are largely predictable on common demarcations that divide the region along east-west and north-south lines — to DataThink partners at Microsoft and the Atlanta Regional Commission (ARC). The work lays the foundation for additional research as well as eventual policy recommendations.
“It is meaningful to know my algorithm and our spatial prediction model has social value,” says Zhang, a senior with a double major in computer science and applied mathematics and statistics. “For me, I was always most interested in the technical part of this work, but the goal always is to benefit society.”
‘Things that really matter’
Such aims rest at the heart of Emory’s university-wide AI.Humanity Initiative, which began last year. The effort is recruiting up to 75 top faculty scholars in artificial intelligence (AI) who will lead interdisciplinary research and education focused on AI innovations to improve the human experience.
Microsoft, which has an established presence in metro Atlanta, began talks with Provost Ravi V. Bellamkonda last year about how they could best engage on the initiative. The company later made a financial gift to support AI innovation.
Bellamkonda designated a portion of the Microsoft donation to the QTM program, which has focused on quantitative humanistic inquiry since it launched in 2014. The department requires students to select a liberal arts discipline to anchor the program’s advanced data science coursework.
“In a future where all the world’s information is at our fingertips, the job of a university is not only to impart knowledge, but to teach students how to communicate, work in teams, deal with ambiguity and complexity and be entrepreneurial in their thinking,” Bellamkonda says.
“QTM’s focus on inquiry-driven problem-solving gives Emory students a tremendous opportunity to gain these skills that will serve them well throughout their careers,” he adds. “In this way, DataThink models the kind of innovative academic experiences we are launching through both the AI.Humanity and Student Flourishing initiatives.”
Clifford Carrubba, the Samuel Candler Dobbs Professor and chair of Emory’s Department of Quantitative Theory and Methods, introduces undergraduates to the department’s summer research project.
The DataThink summer project is the third such practicum for QTM students and the first with collaborators beyond campus. Clifford Carrubba, QTM department chair and Samuel Candler Dobbs Professor, aims to use the rental housing project as a pilot to be scaled up for more projects and more partners.
“Programs like this, hands-on and with real-world implications, prepare our students in a special way that allows them to flourish now and after they graduate,” Carrubba says. “We want to do things that really matter for a lot of people, every day, right now.”
The department had already decided to develop case studies and projects surrounding affordable housing. Tommy Pearce, executive director of Neighborhood Nexus, a partnership between ARC and the Community Foundation for Greater Atlanta, helped narrow the DataThink summer project to rental housing.
The ARC had already subdivided the metro region into segments based on housing sales as part of its Metro Atlanta Housing Strategy. To expand on that work by incorporating renters’ perspectives, it would need to examine whether similar sub-regions existed for rental properties – hundreds of hours of work for a small team.
“As head of a nonprofit, I can say there isn’t enough advanced data work being in our sector or enough data scientists with enough social-issue experience,” says Pearce, who also serves on QTM’s external advisory board. “These students are helping to bridge that gap, moving us along faster on timely issues we want to address.”
Five months’ work in four weeks
The student team — which in addition to Zhang included seniors Billy Ge, Annie Luo, Latifa Tan and Alexia Witthaus — worked full time for four weeks on the project — the equivalent of five months for a full-time researcher. They also attended a series of three technical workshops led by Microsoft, including learning how to incorporate the cloud-computing service Azure in their work.
Microsoft software engineer Almas Myrzatay is a 2018 graduate of Emory College with a double major in economics and computer science. He coordinated the training and paired students with professional mentors.
Microsoft software engineer Almas Myrzatay, an Emory alumnus with a double major in economics and computer science, coordinated the training and paired students with professional mentors. His goal was to create the sort of effort that he would have joined before graduating, just as the QTM program was taking off.
“Microsoft has committed to expanding its engagement within the Atlanta community in general, and Emory will be an active partner for us,” he says. “We have employees who are eager to volunteer, and I want to do everything I can to help build this bridge.”
Even with the assistance, students slogged through real-world challenges, such as unearthing useful data. They settled on using Census tracts as the most accurate source of inputs, says Kevin McAlister, the QTM visiting assistant professor who served as faculty guide with Zhiyung Gong, the QTM director of undergraduate research.
Witthaus, an economics and QTM double major, kicked off the data work by downloading every Census tract in the state, complete with its thousands of characteristics.
Part of the pre-processing work involved becoming familiar enough with that data to decide what characteristics would be important. A key task, for instance, required study of commuter information to decide the project’s geographic borders.
Having decided on a 35-mile radius of downtown Atlanta and 213 characteristics such as median income and unemployment rate for co-variants, Witthaus and the team had to “clean” the data in Python to be ready for input into Zhang’s model.
Students helped each other at every step, constantly narrowing and refining by conducting spatial clustering on the Census tracts until arriving at sub-regions for rental prices similar to the existing housing ownership divisions.
“Our work is mainly descriptive, which lets us see there is a big need for affordable housing, especially in specific neighborhoods,” Witthaus says. “An interesting thing to think about in the future is to combine our analysis with home ownership, to make a policy around housing in general. It’s definitely interesting, even if this is only the beginning.”