Continuous Parameter Pools in Ensemble Differential Evolution
Ensemble of parameters and mutation strategies differential evolution (EPSDE) is an elegant promising optimization framework based on the idea that a pool of mutation and crossover strategies along, with associated pools of parameter settings, can flexibly adapt to a large variety of problems when a simple success based rule is introduced. Modern versions of this scheme successfully attempts to improve upon the original performance at the cost of a high complexity. One of most successful implementations of this algorithmic scheme is the Self-adaptive Ensemble of Parameters and Strategies Differential Evolution (SaEPSDE). This paper operates on the SaEPSDE, reducing its complexity by identifying some algorithmic components that we experimentally show as possibly unnecessary. The result of this de-constructing operation is a novel algorithm implementation, here referred to as "j" Ensemble of Strategies Differential Evolution (jESDE). The proposed implementation is drastically simpler than SaEPSDE as several parts of it have been removed or simplified. Nonetheless, jESDE appears to display a competitive performance, on diverse problems throughout various dimensionality values, with respect to the original EPSDE algorithm, as well as to SaEPSDE and three modern algorithms based on Differential Evolution.
Citation : Iacca, G., Caraffini, F. and Neri, F. (2015) Continuous Parameter Pools in Ensemble Differential Evolution. 2015 IEEE Symposium Series on Computational Intelligence, pp. 1529-1536
ISBN : 9781479975600
Research Group : Centre for Computational Intelligence
Research Institute : Institute of Artificial Intelligence (IAI)
Peer Reviewed : Yes