AD-A256 617 A Massively Distributed Parallel Genetic Algorithm (mdpGA)

The effectiveness of combinatorial search heuristics, such as Genetic Algorithms (GA), is limited by their ability to balance the need for a diverse set of sampling points with the desire to quickly focus search upon potential solutions. One of the methods often used to address this problem is to simulate the theory of punctuated equilibria in the GA. The GA introduced here uses the basic premises derived from punctuated equilibria, but hopes to remedy the problems associated with sudden introduction of new genetic material by relying upon a much greater degree of distribution and an overlapping population architecture. Presented here is a description and preliminary empirical test results of a massively distributed genetic algorithm. On the seventeen test problems attempted, the mdpGA did significantly better than a simple parallel GA. The massive distribution of the GA and the modified population topology yield improvements in speed, and also prove to be far less vulnerable than other genetic algorithms to biases in the function space which lead away from global optima