Xin Jiang, an assistant professor in the Cullen College of Engineering’s Industrial and Systems Engineering (ISE) department, is driven by the rigor and utility of mathematical optimization. Though his background is electrical engineering, his work today in ISE leans more toward the theoretical in a way that helps him to bridge abstract mathematics with real-world applications.
“My research focuses on mathematical optimization,” Jiang said. “Optimization is really about increasing the efficacy of a given goal subject to certain constraints, and I design new patterns for communication and customized algorithms to fix the challenges in these new distribution scenarios. I’m especially drawn to what I call ‘practical theory’: spotting the distinctive structure of real-world problems and leveraging it to develop methods that are both theoretically grounded and demonstrably effective.“
One recent example is distributed optimization, which essentially allows multiple computing devices, or agents, to tackle a problem together, with each device running its own piece of the necessary work to reach the solution. This is an alternative to attempting to solve an algorithmic optimization problem solely through use of a more powerful computer, which may be less cost-effective or perhaps not an option at all.
This approach is especially timely given the impact that recent developments in artificial intelligence have had on computer hardware supplies. Modern AI development depends on both large-scale data and coordinated, high-performance computation infrastructure, so as demand from AI data centers dominates existing and potentially future supplies of RAM and GPU components, distributed approaches may become more necessary than ever before.
“Nowadays, the scale of data is exploding beyond expectations. Due to production times, it’s difficult for hardware like GPUs and memory units to meet the urgent needs of the explosion of big data,” said Jiang. “To me, that means the only remedy is to distribute it.
“My research has been focused on the top level, or software level, which is to design customized distribution algorithms tailored to real-world applications. I’m eager to apply my experience in mathematical distributed optimization to different real-world applications, and now with all of those new opportunities and challenges, I’m interested in collaborating with those researchers with primarily bottom-level or hardware backgrounds so that the entire distribution can be unified — streamlined from the bottom to the top.”
Though his background is in electrical engineering, Jiang invested time into the exploration of different research areas during his Ph.D. studies, and he found that he particularly appreciated the theory and predictability that bely optimization.
“In some areas of engineering, results can be surprisingly context-dependent — there are many interacting factors, and not all of them are easy to observe or control. In contrast, optimization was the most mathematical area that I found: when something works, I know how and why it works, and when something fails, I know that means there is a bug in my work and I need to fix it. I’m very enthusiastic about the process of modeling a messy, real-world problem into a rigorous, abstract mathematical problem,” he said.
“Optimization is everywhere,” Jiang added. “It’s used in many disciplines, from engineering and science to business, and it has a bright future and a wide range of applications. Industrial engineering is a very open-direction field as well, and our department includes many topics within engineering. That breadth is a big reason industrial engineering appeals to me — our department sits at the intersection of many engineering areas and naturally brings together different kinds of problems and perspectives. I’m always hoping to collaborate with people from different backgrounds who can bring their experience from their applied fields to meet mine, and I hope to apply my experience in optimization to help them solve real-world problems and pursue open-ended progress.”