(339j) Optimal Communication Topologies for Particle Swarm Optimization Under Hostile Environments

A class of nature-inspired optimization techniques called swarm intelligence methods involves a systematic exploration and exploitation of the search-space through efficient information exchange between the constituent agents. Such algorithms are common in areas where the optimization problems are inherently difficult due to a lack of complete information about the function landscape, due to the presence of several local minima that could result in premature convergence, and more importantly due to the presence of hostile environments that could result in a partial loss of swarm agents during the exploration phase. Such hostile environments could be seen in applications involving the design of communication channels for efficient information dissemination to a target group, targeted drug-delivery where drug molecules search for the affected site before diffusing, and high-value target localization with a network of drones communicating among themselves. In this work, we study the impact of the loss of agents on the performance of such algorithms as a function of the initial communication topology. We use particle swarm optimization (PSO) to optimize an objective function with multiple sub-optimal regions in a hostile environment and study its performance for a range of network topologies under hostile conditions. The presence of a hostile environment results in changes in the swarm communication topologies during the search-space exploration resulting in varying performances for different initial network configurations. The results reveal interesting relationships between graph-theoretic properties and the algorithmic performance based on which general properties of networks that maximize performance are identified. Moreover, networks with small-world properties are seen to maximize performance under hostile conditions and outperform the standard network topologies. Although the findings of this work are based on performance of the PSO algorithm, the results are fairly generalizable, and should extend to algorithms that rely on efficient communication between agents while searching for an optimal solution.