A Genetic Algorithm With Self-distancing Bits But No Overt Linkage

We present a novel representation and crossover operator for genetic algorithms. Bits are not linked to one another. Instead, the current population suggests a pseudo-distance between each pair of bits; this pseudo-distance really measures the degree to which bits appear to participate in a building block. Then crossover respects the pseudo-distance: it is clumps of nearby bits that have their values copied from parent to child. Thus our new approach does directly (preservation of building blocks under crossover) what other approaches only hope to do indirectly. Our approach is tested on several problems, ranging from simple to very challenging, and the results compared to standard approaches. In these problems, the new approach is successful and usually outperforms the standard approaches.