Improving Evolvability of Genetic Parallel Programming Using Dynamic Sample Weighting

This paper investigates the sample weighting effect on Genetic Parallel Programming (GPP) that evolves parallel programs to solve the training samples captured directly from a real-world system. The distribution of these samples can be extremely biased. Standard GPP assigns equal weights to all samples. It slows down evolution because crowded regions of samples dominate the fitness evaluation and cause premature convergence. This paper compares the performance of four sample weighting (SW) methods, namely, Equal SW (ESW), Class-equal SW (CSW), Static SW (SSW) and Dynamic SW (DSW) on five training sets. Experimental results show that DSW is superior in performance on tested problems.