Evaluating factored evolutionary algorithm performance on binary deceptive functions
Shane Strasser, John W. Sheppard · 2017
Factored Evolutionary Algorithms (FEA) have been shown to be an effective method to optimize objective functions by partitioning the search space into overlapping subpopulations, or factors. FEA is comprised of three main steps: Update, Compete, and Share. While there exists previous work exploring FEA's convergence properties, it is still unknown where FEA obtains most of its performance gains. In this paper, we examine FEA performance by evaluating FEA on a set of commonly used deceptive unitation functions as well as several versions of the Royal Road problem. These problems provide a complex landscape for the search algorithm to explore but are well understood and researched. In evaluating FEA on these problems, we discovered that FEA's Compete step contributes the most to its performance effectiveness. Additionally, we identify a class of problems that may degrade the performance of FEA.