Differential Evolution
Also known as DE
Overview
Differential Evolution (DE) searches by combining candidate solutions through vector-difference mutation, crossover, and selection, making it useful for black-box numerical problems where gradients are unavailable or unreliable. Common variants such as success-history adaptation and L-SHADE adjust control parameters and population size, while current-to-pbest mutation, archives, and multiple-operator selection aim to balance exploration and refinement. Researchers extend DE with constraint-handling and Pareto-ranking methods, late-stage local search, or hybrids with CMA-ES, gradient methods, and reinforcement learning; evaluations often use CEC and BBOB benchmarks as well as engineering and scientific applications. A central challenge is making performance robust across dimensions, constraints, and noisy evaluations without overfitting benchmark suites or sacrificing fast convergence for broader exploration.
Latest papers
23 papers3 papers in the last four weeks, against 2 the four weeks before. 0.0% of all new papers.