High Performance Evolutionary Computation

Abstract

Cost and performance issues have driven many High Performance Computing (HPC) communities to embrace personal computer (PC) clusters. Evolutionary computation (EC) is a global optimization technique that is often used to solve non-linear problems. Because of its computationally intensive nature, EC research is an obvious candidate for HPC utilization. EC applications include diverse fields like schedule optimization, robot navigation, controller design, image processing, discrimination of buried unexploded ordnance (UXO), and others. Our research has focused on genetic programming (GP), one of the many EC subfields. Other researchers using genetic programming techniques have obtained human competitive results, and have patented innovative electronic filter and controller designs. As the EC technique is applied to more complex Department of Defense (DoD) related problems, this research community will require increasing HPC resources. This paper will provide a broad overview of what EC techniques are, with particular emphasis on GP. It will also discuss the impact of EC on HPC resource allocation and utilization. EC and GP techniques are often applied to problems where an incomplete knowledge of the underlying physics exists. These techniques can also be used to solve problems characterized by non-linear mathematics without analytical solutions. We have applied GP techniques to the problems of buried unexploded ordnance discrimination and image noise reduction. Results are encouraging and point to the need for applying EC techniques to the many difficult problems faced by the DoD community. As these techniques gain further acceptance among HPC users, the providers of these computational resources must be prepared to meet the challenge.