Gene Expression Programming with Multi-Threading Evaluation and Gene-Reuse Strategy

As an approach widely used in automatic programming, the efficiency of the traditional GEP algorithm has gradually failed to meet the needs of users since its bottleneck in the evaluation phase. In this paper, a novel strategy named Gene-Reuse is proposed to improve the efficiency of GEP. In contrast to the traditional evaluation phase of GEP, the Gene-Reuse strategy features a novel mechanism that the Gene-Reuse strategy directly reads the pre-saved fitness value of chromosomes if these chromosomes have appeared in the previous population evolution. By applying that mechanism to the traditional GEP, the optimized algorithm can avoid many meaningless repeated calculations that improve the overall efficiency of the algorithm. Further, combining with multi-threading technology, a new Gene Expression Programming algorithm MTEGR-GEP that has significant performance compared with the traditional GEP algorithm is introduced to solve the existing problems of GEP mentioned above. Experimental results on several symbolic regression problems show that MTEGR-GEP has a significant improvement in efficiency compared to the traditional GEP.