A Stage-Specific Gene Expression Framework for Promoting Genetic Algorithm

Abstract

Genetic algorithm (GA) has become very popular for tackling computationally expensive numerical optimization problems. To optimize their performance and convergence rate, we propose and demonstrate an embedded stage-specific gene expression framework (SGEF) with a two-tier coding scheme and a filter operator based on variational auto-encoder. For the first time, the structure of variational autoencoder is used to map the parent population of a genetic algorithm from the uniformly distributed high dimensions to the Gaussian distributed low dimensions. The design of the filter operator which can maintain the trade-off balance between exploitation and exploration is also analyzed in detail. The experimental results demonstrated the good performance of the proposed SGEF in promoting both single-objective optimization genetic algorithms and multiple-objective optimization genetic algorithms.