RDEx-CASK: Enhanced Evolutionary Algorithm for Constrained Optimization
Researchers have introduced RDEx-CASK, an advanced extension of the RDEx-CSOP evolutionary algorithm designed to address stagnation and late-stage variance in constrained single-objective optimization problems. The study, published on arXiv, details three primary modifications: independent sampling of the second scale factor from a truncated Cauchy distribution, the integration of a small feasible-only JADE-style archive, and a per-individual stagnation counter. This counter triggers specific local overrides after 180 generations without improvement, such as pulling solutions toward the global best and adjusting sampling parameters. Experimental results on the CEC CSOP suite demonstrate that RDEx-CASK competes effectively with state-of-the-art algorithms like RDEx, UDE-III, and CL-SRDE. Notably, it shows improved time-to-target performance across most test problems while maintaining high feasibility-aware solution quality. This development represents a significant technical advancement in neural and evolutionary computing, offering enhanced efficiency for complex optimization tasks.
Wire timeline
RDEx-CASK: Enhanced Evolutionary Algorithm for Constrained Optimization
Researchers have introduced RDEx-CASK, an advanced extension of the RDEx-CSOP evolutionary algorithm designed to address stagnation and late-stage variance in constrained single-objective optimization problems. The study, published on arXiv, details three primary modifications: independent sampling of the second scale factor from a truncated Cauchy distribution, the integration of a small feasible-only JADE-style archive, and a per-individual stagnation counter. This counter triggers specific local overrides after 180 generations without improvement, such as pulling solutions toward the global best and adjusting sampling parameters. Experimental results on the CEC CSOP suite demonstrate that RDEx-CASK competes effectively with state-of-the-art algorithms like RDEx, UDE-III, and CL-SRDE. Notably, it shows improved time-to-target performance across most test problems while maintaining high feasibility-aware solution quality. This development represents a significant technical advancement in neural and evolutionary computing, offering enhanced efficiency for complex optimization tasks.
cs.AI updates on arXiv.org