At Cray we’re always looking way down the road … years, even decades into the future. We’re not developing products for next quarter. We’re developing products for questions our customers might not even know they have yet. That’s how high-performance computing works.
So as you can imagine, we pay very close attention to what’s coming … and that includes the next generation of computer scientists and engineers. These individuals are going to be the ones shouldering an awesome responsibility in the coming decades as big data gets bigger, artificial intelligence flexes its muscles more and more, and problems grow in complexity.
Jeffrey Dudek is one of these next-generation computer scientists. He’s a third-year Ph.D. student in the Department of Computer Science at Rice University and doing some really exciting work with machine learning, probabilistic inference and SAT solving. He’s also the winner of the 2017 Ken Kennedy Institute – Cray Inc. Graduate Fellowship. Ken Kennedy was a computer scientist and software pioneer who served on Cray’s board of directors for many years. The fellowship named in his honor and endowed by Cray provides support to graduate students involved in HPC-related work.
We met Jeffrey recently and got a look into the future of computer science research. What we saw gives us a lot of reason for optimism.
Becoming a Computer Scientist
Dudek didn’t start his university career with computer science as an end goal. He’d always felt a strong pull toward mathematics and did a dual math/computer science major as an undergraduate. But when it came time to decide between the two for his PhD program, computer science won.
“Math does appeal to me,” he says. “But I like to do it in service of some application. I went with computer science because I could be closer to the interesting problems that people care about.” Internships with Microsoft and Google, and a great advisor in computational engineering professor Dr. Moshe Vardi reinforced his choice.
With the practical application goal in mind, Dudek has focused his research on aspects of the Boolean-Satisfaction Problem (SAT). SAT is one of the most fundamental problems in computer science and has application to many fields, including artificial intelligence, programming languages, biology and more.
Dudek’s master’s thesis, “Random CNF-XOR Formulas,” explores hashing-based approaches for the related problems of constrained sampling and constrained counting. These problems have broad application in areas such as probabilistic reasoning, machine learning, statistical physics and others. In his research, Dudek analyzes the use of SAT solvers to solve CNF-XOR formulas as part of hashing-based algorithms, and presents a first study of random CNF-XOR formulas. (The thesis isn’t the first work to use SAT solvers to solve CNF-XOR formulas for constrained counting, but it is the first work to theoretically analyze their performance.)
True to his word, the work has a very tangible benefit. “We’re working to apply this research to estimating the failure rate of power networks in the event of a natural disaster,” Dudek says. These estimates can be computed through constrained counting and, ultimately, civil engineers can use the knowledge to design more resilient networks.
But it’s more than practical application that drives someone to dedicate years to the pursuit of a single area of study. For Dudek, it’s fun. “The specifications of these constrained problems are very clean. The algorithms and techniques end up being elegant, clean and fun to work with.”
With his Ken Kennedy Institute – Cray Inc. fellowship Dudek will present a poster at the 2018 Rice Oil and Gas HPC Conference and — as an added bonus as he likes to travel — will attend the premier international gathering of AI researchers, the International Joint Conference on Artificial Intelligence in Sweden.