Thursday, Oct. 01, 2009
In Utah State University computer scientist Daniel Bryce’s classroom, students use two objects often found in children’s toy boxes to learn about sophisticated planning concepts: a slider puzzle and a Rubik’s Cube.
“Where toys become serious research problems is when the puzzles are solved with less than perfect information,” Bryce says. “Imagine solving a slider puzzle or a Rubik’s cube with closed eyes.”
With each slide of a tile or twist of the cube, the seeker accumulates knowledge, chips away at uncertainty and is one step closer to a solution, Bryce says.
“Planning with partial information is a familiar activity in an everyday person’s life, but imbuing intelligent systems with this capability is a formidable challenge,” he said.
The assistant professor, who earned a doctorate from Arizona State University in 2007, joined USU’s Department of Computer Science in August 2008. His dissertation, “Scalable Planning Under Uncertainty,” received the 2009 Best Dissertation Award at the International Conference on Automated Planning and Scheduling held Sept. 19-23 in Thessaloniki, Greece.
As part of the award, Bryce received funding to travel to the gathering, where he presented his research and was formally recognized for his achievement.
“Dr. Bryce received the award for his research on heuristics to speed up planning under uncertainty,” says Sven Koenig, a senior member of the ICAPS program committee. “On behalf of the ICAPS Council, conference organizers and the award committee, I offer congratulations on a job well done.”
Many heuristics — or techniques designed to solve problems — have been developed to tackle simple challenges but not all are scalable large, complex problems, Bryce says.
“Most current approaches to planning under uncertainty are not scalable,” he says. “They simply focus on small problems where the challenge is finding an optimal solution. I study contingency planning in large problems where just finding any solution is the challenge.”
The idea, Bryce says, is to simplify the structure of the problem and thereby find a close-to-correct solution quickly.
“Solutions to simplified problems — heuristics — inform techniques for solving the actual problem,” he says. “The magic is in designing simplifications that work on a variety of problems.”
Applications for which Bryce’s heuristics are designed range from robotics and military surveillance to genomics and cancer research. He’s currently collaborating with USU associate professor Vladimir Kulyukin, who is developing a hand-held guidance tool for people with visual impairments.
“If you consider the path between USU’s Old Main and the Merrill-Cazier Library, you can envision a straight line,” Bryce says. “But when you actually walk the path, you’ll encounter a number of obstacles — trees, benches, uneven pavement — that cause you to deviate from a straight journey. With our system, we have to consider how people who are unable to see will deal with these challenges to successfully reach their destination.”
The department’s undergraduate research coordinator, Bryce is pursuing a variety of projects involving artificial intelligence that span topics of planning, Markov decision processes, heuristic search, multi-criteria decision making and knowledge-based learning.
In addition to automated planning with uncertainty, his current efforts include applications of AI to systems of biology, along with the Bootstrapped Learning Project, sponsored by the Defense Advanced Research Projects Agency of the U.S. Department of Defense, for human instruction of field-programmable intelligent systems.
Writer: Mary-Ann Muffoletto, 435-797-3517, firstname.lastname@example.org