Evolution of a Metaheuristic for Aggregating Wisdom from Artificial Crowds

Abstract

Approximation algorithms are often employed on hard optimization problems due to the vastness of the search spaces. Many approximation methods, such as evolutionary search, are often indeterminate and tend to converge to solutions that vary with each search attempt. If multiple search instances are executed, then the wisdom among the crowd of stochastic outcomes can be exploited by aggregating them to form a new solution that surpasses any individual result. Wisdom of artificial crowds (WoAC), which is inspired by the wisdom of crowds phenomenon, is a post-processing metaheuristic that performs this function. The aggregation method of WoAC is instrumental in producing results that consistently outperform the best individual. This paper extends the contributions of existing work on WoAC by investigating the performance of several aggregation methods. Specifically, existing and newly proposed WoAC aggregation methods are used to synthesize parallel genetic algorithm (GA) searches on a series of traveling salesman problems (TSPs), and the performance of each approach is compared. Our proposed method of weighting the input of crowd members and incrementally increasing the crowd size is shown to improve the chances of finding a solution that is superior to the best individual solution by 51% when compared to previous methods.