PruneSymNet: A Symbolic Neural Network and Pruning Algorithm for Symbolic Regression

• DOI: 10.48550/arxiv.2401.15103
• Publication Date: 2024-01-25
• Authors: Min Wu; Weijun Li; Lina Yu; Wenqiang Li; Jingyi Liu; Yanjie Li; Meilan Hao
• Keywords: Pruning; The Symbolic; Symbolic Regression; Symbolic trajectory evaluation

Symbolic regression aims to derive interpretable symbolic expressions from data in order better understand and interpret data. %which plays an important role knowledge discovery machine learning. In this study, a network called PruneSymNet is proposed for regression. This novel neural whose activation function consists of common elementary functions operators. The whole differentiable can be trained by gradient descent method. Each subnetwork the corresponds expression, our goal extract such subnetworks get desired expression. Therefore, greedy pruning algorithm prune into while ensuring accuracy fitting. preserves edge with least loss each pruning, but often not optimal solution. alleviate problem, we combine beam search during obtain multiple candidate time, finally select expression smallest as final result. It was tested on public set compared current popular algorithms. results showed that had accuracy.