An Enhanced Scale-Free Differential Evolution for Dendritic Neural Network Training

The parameter training of machine learning models imposes strict requirements on the optimization performance of learning algorithms. However, as application scenarios grow increasingly complex, conventional learning algorithms, such as back-propagation and least squares methods, can no longer keep pace with the rising complexity of modern machine learning models. Given the superior performance of evolutionary algorithms in solving nonlinear, constrained, and large-scale optimization problems, a growing body of research has explored the application of evolutionary algorithms to the parameter training of machine learning models. Notably, the optimization process of evolutionary algorithms is less sensitive to the specific information of the search space, allowing them to effectively address the non-interpretability of machine learning models. In this paper, we propose a differential evolution algorithm with a scale-free network-based mutation operator and apply it to a type of machine learning model, namely the dendritic neural network, as well as its variants. When compared with nine state-of-the-art evolutionary algorithms, the proposed algorithm exhibits significant superiority in training performance. This result demonstrates that our algorithm possesses better optimization capability and stronger robustness.