IfA Student Wins Harvard Computer Competition
by Chris Beaumont, IfA Graduate Student
Chris Beaumont (right) and his project partner Blessing Okeke won the IACS Computational Challenge. Harvard SEAS.
Academic research occasionally gets a reputation for being impractical. As astronomers, we sometimes respond to this by emphasizing the practical aspects of our work, such as our searches for potentially dangerous near-Earth asteroids. However, if these were our only justifications for doing research, there would be more efficient ways of achieving our goals. After all, most of us don’t look for asteroids, and most of our research focuses on esoteric questions that will not percolate back to the larger community. I think we can better justify our work by appealing to the unique culture of open-ended exploration and collaboration that academia encourages. Academia is unique in its ability to incubate and cross-pollinate ideas from diverse fields. I recently experienced this firsthand.
For the past year, I have been carrying out my PhD research in Cambridge, Mass., where I collaborate with researchers at Harvard. This January, I participated in a competition to develop computational strategies for responding to natural disasters. The competition organizers—professors who study disaster relief at the Harvard Institute for Applied Computational Science (IACS)—presented us with a database of the roads in Cambridge. These roads had been littered with debris, simulating the aftermath of a major hurricane. This debris rendered the roads impassable, and cut off aid workers from sick and injured residents.
Our goal was to develop a strategy for clearing these roads in as efficient manner as possible, given limited resources. On each simulated day, we were given a certain number of trucks to clear roads. These trucks had to clear their way into the city, opening access to as many people as possible. At regular intervals, we were penalized for the number of residents still without access to an open road. This mimics real-world considerations, where common health concerns like dehydration and cholera progress on timescales of hours to days.
We had to synthesize all of this information—the finite resources, the layout of roads and debris, the varying population density throughout the city—into a strategy for clearing debris. We had a week to design our algorithm, and three hours of computing time on one of Harvard’s supercomputers to come up with a solution.
I realized that this problem, while seemingly unrelated to astronomy, overlaps with the problems I solve every day in my research. For example, the idea of finding a good strategy for clearing debris is similar to searching for a good model for describing astronomical data—the underlying computation is the same.
I decided to use an approach called simulated annealing. This method tries to improve upon an initial, arbitrary road-clearing solution by considering whether slight modifications to the plan lead to a better outcome. Repeating this process millions of times leads to a very efficient road-clearing plan. I knew about simulated annealing because I had used it to design a strategy for Pan-STARRS to take repeated images of the sky in such a way that the regions it misses in a single image (due to gaps in the camera’s huge array of detectors) are filled in by subsequent images.
This strategy turned out to be highly effective. Not only did it produce the winning solution, but it outperformed the solution designed by the contest organizers. If the simulated disaster were a real emergency, this difference could easily translate to dozens or hundreds of saved lives.
It is hard to imagine this situation occurring outside of academia. I would not have participated in the contest if my PhD advisor had not encouraged me to collaborate with researchers at Harvard. I would not have known about simulated annealing had it not been for my work with Pan-STARRS. And the competition would not have been held if it weren’t for the newly commissioned IACS, set up to encourage collaboration between engineers and physical scientists. Components of academia can indeed be impractical. But the ideas generated in this unstructured environment can trickle out into the larger world in surprising ways.
In addition to winning the IACS Computational Challenge, Chris Beaumont was also named 2012 ARCS Scholar of the Year by the Honolulu Chapter of the Achievement Rewards for College Scientists and received the UH 2012 Student Excellence in Research Award at the doctoral level.