LINKGAUGE: tackling hard deceptive problems with a new linkage learning genetic algorithm
Miguel Nicolau, Conor Ryan · 2002
A novel approach to obtaining a tight linkage between genes in a genetic algorithm is described, and a new system based on that approach, LINKGAUGE, is proposed. Experiments presented draw a comparison between the standard messy genetic algorithm and LINKGAUGE, and show that the latter avoids deceptive traps and early convergence, with minimal computational cost. The scalability potential of the new approach is illustrated with results for two hard deceptive problems.