Empirical Assessment of Human Learning Principles Inspired PSO Algorithms on Continuous Black-Box Optimization Testbed

This paper benchmarks the performance of one of the recent research directions in the performance improvement of particle swarm optimization algorithm; human learning principles inspired PSO variants. This article discusses and provides performance comparison of nine different PSO variants. The Comparing Continuous Optimizers (COCO) methodology has been adopted in comparing these variants on the noiseless BBOB testbed, providing useful insight regarding their relative efficiency and effectiveness. This study provides the research community a comprehensive account of suitability of a PSO variant in solving selective class of problems under different budget settings. Further, certain rectifications/ extensions have also been suggested for the selected PSO variants for possible performance enhancement. Overall, it has been observed that SL-PSO and MePSO are most suited for expensive and moderate budget settings respectively. Further, iSRPSO and TPLPSO have provided better solutions under cheap budget settings where iSRPSO has shown robust behaviour (better solutions over dimensions). We hope this paper would mark a milestone in assessing the human learning principles inspired PSO algorithms and used as a baseline for performance comparison.