Sign Change Detection Based Fitness Evaluation for Automatic Implicit Equation Discovery

Jiahao Wen,Junlan Dong,Jinghui Zhong

DOI: https://doi.org/10.1145/3638529.3654015

2024-01-01

Abstract:Automatic implicit equation discovery is a meaningful and challenging problem in symbolic regression. The current common methods for the automatic discovery of implicit equations include derivative calculations and comprehensive learning. However, both methods come with their own set of challenges. Derivative calculations pose difficulties in handling sparse data. The comprehensive learning method may encounter problems associated with multiple multiplications, making it difficult to find the optimal equation. Inspired by Bolzano's theorem,we propose a new evaluation mechanism known as the "Sign Change Detection (SCD) based Fitness Evaluation". The main idea of our proposed mechanism is to approximate the solution of an equation using Bolzano's theorem. This mechanism can overcome the limitations associated with derivative calculations and comprehensive learning methods. Furthermore, we integrate this mechanism with self-learning gene expression programming (SL-GEP) to develop a new SCD-GEP method. Experimental results have shown that the proposed method surpasses the compared approaches in discovering implicit equations, achieving a higher success rate in finding optimal solutions.