A Deep Learning Assisted Gene Expression Programming Framework for Symbolic Regression Problems

Jinghui Zhong, Yusen Lin, Chengyu Lu, Zhixing Huang

Abstract: Genetic programming is a powerful evolutionary algorithm that solves user-defined tasks through the evolution of computer programs. Selecting a proper set of function primitives is a fundamental and challenging operation in applying GP to real applications. Traditional manual design methods require a lot of domain knowledge and are not effective and convenient enough. To address this issue, this paper proposed an automatic function primitive identification mechanism. The key idea is to train a deep convolutional neural network to predict the probability of the existence of a function primitive in the target solution. During the evolution of GP, function primitives with higher probabilities are more likely to be selected to construct solutions. The proposed method is tested on nine benchmark problems and the experimental results have demonstrated the efficacy of the proposed method.