“I like working on problems that are high risk and high reward,” said Kuldeep Meel, a fourth-year Ph.D. student in computer science at Rice University and winner of a 2016 IBM Ph.D. Fellowship.
Meel’s research ranges from artificial intelligence (AI) to parallel computing. “Some problems are easily understood and resolved,” he said, “but I really love the problems we haven’t yet solved after two or three years.” Meel is mentored by Moshe Vardi, Rice’s Karen Ostrum George Distinguished Service Professor in Computational Engineering and Director of the Ken Kennedy Institute for Information Technology.
Rice’s small size has permitted Meel to work on interesting collaborations in such fields as game theory and parallel computing, “unrelated to my primary doctoral dissertation research,” he said, “but the CS department is a hotbed for collaborations initiated by students working in entirely different research areas. All these collaborations resulted from hallway discussions or interactions in social settings.”
His recent IBM award is linked to his work on constrained sampling and counting in high-dimensional spaces, two fundamental problems in AI with applications in probabilistic reasoning and computer-aided verification. “The primary contribution of my work,” he said, “is a novel hashing-based approach that not only provides strong theoretical guarantees but also scales well to large spaces, an important requirement for the design of reliable automated reasoning software and hardware systems.”
Meel said his doctoral research has focused on counting high-dimension combinatorial spaces, a key aspect of automated reasoning techniques. “I love the simplicity of the statement of the problem,” he said, and gave a counting example.
To count the number of coffee drinkers in a small group, the surveyor might visit every office in the group. But to scale up, for example to count the number of coffee drinkers in Houston or the U.S., it would be hard to count every home or office. By dividing people into small groups with a roughly proportional number of coffee drinkers, the number of people in every group would be small enough to count with precision.
“What we have shown over the years using deep theoretical analysis,” he said, “is that it is possible to `efficiently’ partition people without assuming anything about their distribution, and my current work focuses on extending the framework to a broader range of applications.”
High-risk, high-reward research excites Meel, but he is also driven by results with a significant impact. “The simplicity of the problem but the requirement of deep theoretical analysis–along with its impact on an increasingly connected world–is the key to my motivation,” he said.