Notice of Online Archive

  • This page is no longer being updated and remains online for informational and historical purposes only. The information is accurate as of the last page update.

    For questions about page contents, contact the Communications Division.

Biology and math might not appear to be interrelated fields of study, but Konstantinos Bousmalis ’05 (Thessanoliki, Greece) found a quicker way to solve some of the world’s toughest problems by combining the two this summer.

Working with Jeffrey Pfaffmann, assistant professor of computer science, Bousmalis worked on making it easier for an evolutionary algorithm to find the best possible solution to complex quandaries. Evolutionary algorithms, which are mathematical processes rooted in theories of natural selection, find good solutions to complex, time-consuming questions, Pfaffmann says.

Bousmalis and Pfaffmann collaborated on a paper outlining a new method they developed for the evolutionary algorithm and presented their research June 20-23 at the Congress on Evolutionary Computation in Portland, Ore. The conference brings together top researchers, practitioners, and students from all over the world to discuss advances in evolutionary computation.

“We [tried] to find the most sufficient solution to the problem,” says Bousmalis, a double major in computer science and mathematics.

The pair collaborated through Lafayette’s distinctive EXCEL Scholars program, in which students conduct research with faculty while earning a stipend. The program has helped to make Lafayette a national leader in undergraduate research. Many of the more than 160 students who participate each year share their work through articles in academic journals and/or conference presentations.

Like natural selection, in which only those species best able to cope in an environment survive, evolutionary algorithms find answers to problems by being modified until the best possible result is produced, Pfaffmann says. At some point, however, the evolutionary algorithms reach the point where they cannot be tweaked anymore and get stuck in complex problems.

“What we [did] is use this (new) method in particular to the evolutionary process to help the evolutionary algorithm get untrapped,” says Bousmalis.

“Think of it as a landscape and you’re the evolutionary algorithm crawling through the landscape randomly making choices,” Pfaffmann says. “You get to the top of a mountain and you think this is as good as it gets because I just can’t get anywhere.

“What we wanted to do is add in a component that will keep track of how much a certain area has been sampled. So what will happen is, if you’ve been on your peak for a really long time and you’re getting really bored with it, it’s time to get bigger and bigger jumps that will put you into a better region. You take that jump by making the modifications bigger and bigger based on how well you know your territory.”

The new method could mean significant changes in the way scientists solve problems.

“It’s more efficient in the sense that it takes fewer iterations to find the correct point on average,” Bousmalis says. “There are a series of problems that are either unsolvable or too time-consuming to solve. Let’s say you have a problem that takes a couple of days to find a solution with the traditional evolutionary algorithm. By using ‘scouting’ you reduce the time significantly.”

More than solving unsolvable problems, the new method could revolutionize the way scientists think about complex systems.

“Say you’re designing an aircraft and you’re trying to make a more effective wing,” Pfaffmann says. “Engineers have a good idea of where the wing should be, but maybe they’re not quite seeing everything and there’s a solution not within their sight — it might be a radically different solution. So then you might end up with an airplane wing that looks more like a bird’s wing.”

Bousmalis ran experiments, analyzed the results, and worked with those results to get a better solution.

“He is fairly independent, so it [led] to some different results than what I would have maybe thought of,” says Pfaffmann, who chose Bousmalis based on his interest in this type of research and his knowledge of artificial intelligence and biological evolution.

Regardless of whether Bousmalis chooses a research-related career, the experience will be valuable, Pfaffmann says.

“[It’s taught] him something about research — not only what happens if something does work out, but the most effective way to make it work out.”

Bousmalis says his work with Pfaffmann enhanced his education in a way classroom work could not.

“First of all, I have experience in doing research, which is what I would like to do the rest of my life,” he says. “And I’m interested in biology and how you can use biological concepts, in a sense, in a computer science context.”

Especially this particular concept, Bousmalis says.

“It’s very exciting to contribute to something like this,” he says. “The particular concept, it’s so creative, it’s not strictly mathematical in a sense, it’s artistic in a sense, like finding an improvement.”

Which is precisely the objective of the new method.

“One of the goals of these evolutionary techniques is to find things that are more creative, that aren’t necessarily obvious,” Pfaffman says.

In a Software Engineering class, Bousmalis helped create a basic version of The SIMS, a game in which users manipulate groups of simulated people in various scenarios and watch the effects of their input. Last fall, he participated in the annual Association of Computing Machinery Mid-Atlantic Programming Contest, in which Lafayette won the University of Virginia site competition and placed fifth overall among 73 institutions.

As a first-year student, he was among a dozen students representing Greek members of the European Parliament in the annual European Union Simulation sponsored by the European American Institute. Bousmalis is a member of the International Students Association and has participated in several mathematics competitions at Lafayette.

As a national leader in undergraduate research, Lafayette sends one of the largest contingents to the National Conference on Undergraduate Research each year. Forty-two students were accepted to present their work at the last annual conference in April.

Categorized in: Academic News