Caleb Ju, a Ph.D. student in the H. Milton Stewart School of Industrial and Systems Engineering (ISyE) has been awarded a highly competitive Department of Energy Computational Science Graduate Fellowship (DOE CSGF). The DOE CSGF is a four-year-long fellowship that provides robust benefits and opportunities to students pursuing doctoral degrees in fields that use high-performance computing (HPC) to solve complex science and engineering problems. 

Ju’s research interest primarily focuses on creating scalable optimization algorithms with applications to problems such as decision making under uncertainty.

“Since I am interested in both theory and practice, I specifically applied to the math and computer science track of the fellowship,” said Ju. "I enjoy designing algorithms from scratch, analyzing their theoretical properties, and applying those results to software to solve real-world problems."

Most of Ju’s application discussed research from his time spent as an undergraduate student at the University of Illinois at Urbana-Champaign; he worked to solve large-scale graph problems – some of which included over 200 million edges – using a unique method.

“Instead of solving graph problems using a discrete method, we solved it using continuous optimization, specifically linear programming,” Ju elaborated. “One advantage of our approach is generality. While general solutions are typically slower, our algorithm with an efficient line search becomes competitive with some of the fastest discrete methods. This means users of our software can get the same performance as specialized codes, without needing to spend months developing new algorithms and software.”

While the award emphasizes HPC research, Ju is a nontraditional winner of the fellowship. He originally applied for the funds while a graduate student in Georgia Tech’s College of Computing, planning to research computer science with HPC applications. He has since transferred to ISyE, where he will be advised by A. Russell Chandler III Associate Professor George Lan, with a focus on operations research.

“I often found the algorithms used to solve an optimization problem were not well-suited for today’s supercomputers due to large data movement between processors, lack of parallelism, and so forth,” said Ju. “By transferring to ISyE, I will be able to combine my training in mathematical modeling and optimization, as well as my background in computer science, to design new algorithms cognizant of modern computer architectures to achieve better performance.”

Lan said, “My warmest congratulations go out to Caleb for winning this highly competitive award. With the support of DOE CSFG, Caleb will work with me on the design of efficient dynamic stochastic optimization algorithms that can both exploit problem structures and utilize HPC in an effective manner.”

While he is interested in optimization, generally, Ju also keeps in mind possible applications for his work. One example of this is reinforcement learning.

“In this area, with the use of fast optimization algorithms, machines can outperform humans in video games,” he explained.” I am looking to extend these results to help solve fundamental scientific and engineering problems.

“I try to keep in mind what the application will be, and whether it will be useful to a scientist or engineer,” Ju added.

The original version of this news release was written by Kristen Perez.