ACM TechNews

Could Evolutionary Computation Cut Billions of Years in Solving Problems?

Hampshire College (07/30/08) Thomas, Elaine

Researchers from Hampshire College in Massachusetts and the State University of New York have used evolutionary computation techniques to solve a century-old algebra problem faster and more efficiently than previous efforts. Evolutionary computing is an increasingly popular sub-field of computer science in which evolutionary processes are built into computer software. The user selects the elements that will be used and how the desirability of particular designs can be measured. The system then creates and tests random combinations of the chosen elements, with better combinations being allowed to create offspring. After many generations, this evolutionary process often produces novel and useful designs and inventions. The researchers used a Beowulf-style computer cluster to simulate the Darwinian processes to look for discriminator terms, majority terms, and Mal'cev terms for finite algebras. Success required finding a formula of manageable size produced in a reasonable amount of time. Two methods for solving the problem had previously been developed, but neither one met both requirements. Evolutionary computation was able to produce useful formulas of fewer than 300 characters in just a few hours.

http://www.hampshire.edu/8693.htm

---

© Copyright 2008 Information, Inc. This service may be reproduced for internal distribution.