Automatic generation of evolutionary operators
Vinícius Veloso de Melo, Grazieli Luiza Costa Carosio · 2013
Genetic Programming (GP) is presented as an approach for the automatic generation of high quality mutation strategies for the Differential Evolution (DE) algorithm. To evaluate the generated mutation strategies, some well known continuous global optimization problems were selected. Also, a multi-layer perceptron neural network was trained for regression and classification tasks. Statistical analysis of the results showed quite satisfactory performance of the discovered mutation strategy (dubbed ADAM), presenting competitive or better results than classical mutation strategies. The contribution of this work is to show that GP is a potential tool for exploring new ideas in metaheuristics development, allowing for an automatic generation of mutation strategies that can present fast performance and high quality solutions.