Delta Coding: An Iterative Search Strategy for Genetic Algorithms

A new search strategy for genetic algorithms is introduced which allows iterative searches with complete reinitialization of the population preserving the progress already made toward solving an optimization task. Delta coding is a simple search strategy based on the idea that the encoding used by a genetic algorithm can express a distance away from some previous partial solution. Delta values are added to a partial solution before evaluating the tness; the delta encoding forms a new hypercube of equal or smaller size that is constructed around the most recent partial solution. Results are presented on two optimization problems involving geometric transformations; solving these problems with precision is diicult for conventional genetic algorithms as well as traditional mathematical optimization techniques. Tests using single population and distributed genetic algorithms are compared to delta coding. Delta coding is shown to produce more precise solutions while reducing the amount of work necessary to reach the solution.