Evolutionary Canonical Particle Swarm Optimizer --- A Proposal of Meta-optimization in Model Selection

Authors:
Hong Zhang;Masumi Ishikawa
Affiliations:
Department of Brain Science & Engineering, Kyushu Institute of Technology, , Japan 808-0196 and Department of Brain Science and Engineering, Graduate School of Life Science & Systems Engineering, ...;Department of Brain Science & Engineering, Kyushu Institute of Technology, , Japan 808-0196
Venue:
ICANN '08 Proceedings of the 18th international conference on Artificial Neural Networks, Part I
Year:
2008

Citing 4
Cited 0

Recent approaches to global optimization problems through Particle Swarm Optimization

Natural Computing: an international journal
Dissipative particle swarm optimization

CEC '02 Proceedings of the Evolutionary Computation on 2002. CEC '02. Proceedings of the 2002 Congress - Volume 02
Designing Particle Swarm Optimization: performance comparison of two temporally cumulative fitness functions in EPSO

AIA '08 Proceedings of the 26th IASTED International Conference on Artificial Intelligence and Applications
A hierarchical particle swarm optimizer and its adaptive variant

IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics

Quantified Score

Hi-index	0.01

Visualization

Abstract

We proposed Evolutionary Particle Swarm Optimization (EPSO) which provides a new paradigm of meta-optimization for model selection in swarm intelligence. In this paper, we extend the technique of online evolutionary computation of EPSO to Canonical Particle Swarm Optimizer (CPSO), and propose Evolutionary Canonical Particle Swarm Optimizer (ECPSO) for optimizing CPSO. In order to effectually evaluate the performance of CPSO, a temporally cumulative fitness function of the best particle is adopted in ECPSO as the behavioral representative for entire swarm. Applications of the proposed method to a suite of 5-dimensional benchmark problems well demonstrate the effectiveness. Our experimental results clearly indicate that (1) the proper parameter sets in CPSO for solving various optimization problems are not unique; (2) the values of parameters in them are quite different from that of the original CPSO; (3) the search performance of the optimized CPSO is superior to that of the original CPSO, and to that of RGA/E except for the result to the Rastrigin's benchmark problem.