Improved Evolutionary Algorithm Performance
through Graph Based Evolutionary Algorithms

Sunday, 1:30 - 4:30 p.m.

(Pavilion Suite 2)

Dr. Kenneth Mark Bryden and Dr. Daniel Ashlock, Iowa State University

Overview: Optimization techniques that search a solution space without designer intervention have become important tools in the engineering of many thermal fluid systems. Evolutionary algorithms (EAs) are among the most robust of these optimization methods because the ability to optimize many designs simultaneously makes EAs less susceptible to premature convergence than comparable gradient search methods.

A recently developed evolutionary optimization techniques, graph based evolutionary algorithms (GBEA), utilizes population graphing to impose a topology or geography on the evolving solution set. In many cases in nature, the ability of a particular member of a population to mate and reproduce with another is limited. The factors creating these limits vary widely and include geographical distance, mating rituals, and others. The effect of limiting the potential mating pool is a reduced rate of transmission of genetic characteristics and an increased diversity within the populations. A combinatorial graph (or graph) G is a collection V(G). The vertices and E(G) of edges where E(G) is a set of unordered pairs from V(G). The vertices will contain configurations from the evolving population and the edges will designate pairs of vertices that are adjacent, so that reproduction and crossover may take place between them. By utilizing a graph to impose a geography on the mating population, limits to mating analogous to those observed in nature are created.


Current studies have found that the choice of graph in a GBEA can affect the number of mating events required to solve a problem by as much as 12-fold. Also problems with simpler fitness landscapes performed best with highly connected graphs (10 times faster) with less connected graphs.

This workshop will focus on 1) teaching the participants how to implement graph based evolutionary algorithms within their work, 2) developing an understanding of how the rate of mixing and rate of information spread within an evolving population affect the solution time and diversity, and 3) describe how a GBEA can be implemented to tune an evolving population to a problem landscape. Additionally, this workshop will present new methodologies to achieve hypermixing within an evolving population and discuss interactive visualization of information spread within the working population of solutions.

Instructors' Background: Dr. Kenneth “Mark” Bryden is an assistant professor of Mechanical Engineering at Iowa State University and a senior research associate of the Virtual Reality Applications Center. Dr. Bryden’s primary research interests are in the area of virtual engineering, which requires the integration of virtual reality, high performance computing and new computational algorithms to solve complex, tightly coupled engineering and decision analysis problems. The goal of this research is to develop interactive user-centered technical decision-support tools. Key among these tools is the development of robust optimization methods for complex engineering systems. His work in this area currently includes development of 1) a taxonomy of engineering optimization and design problems, 2) graph based evolutionary algorithms, 3) planned tournament selection for evolutionary algorithms, and 4) computational algorithms for analytical solution of differential equations.

Dr. Daniel Ashlock is an Associate Professor of Mathematics at Iowa State University and currently serves as the Chair of the Complex Adaptive Systems graduate program. He is also on the faculty of the Bioinformatics and Computational Biology Program and the Human Computer Interface program at Iowa State and is a faculty associate of the Virtual Reality Applications Center. Dr. Ashlock’s research is in several areas including machine learning, bioinformatics, evolutionary computation, and theoretical biology. His current main research areas include improving methodology in evolutionary computation and mining biological processes for algorithmic innovations.