Symmetric Comparator Pairs In The Initialization Of Genetic Algorithm Populations For Sorting Networks

This paper concerns a well-known combinatorial optimization problem called sorting networks, which has been of some interest to the computer science and engineering community for many decades. A simple method for the initialization of genetic algorithm (GA) populations for the sorting network problem is presented in which networks are constructed with a structural symmetry, making use of symmetric comparator pairs. This method, and several variations thereof, are motivated and compared experimentally with a more standard initialization method. The benefits of symmetric comparator pairs on sorting networks, both before and after GA evolution, are demonstrated. Some interesting but unexplained observations regarding the distribution of such networks are presented as well.