A Study of Dynamic Neural Network Framework Based on Data Preprocessing and PDE Integration

Deep Learning is one of the promising branches of Artificial Intelligence that has made milestones in various aspects. The widely used Convolutional Neural Network (CNN) approach requires complex model structures to analyze data. However, this complexity reduces the interpretability of the model and may cause problems of overfitting and decreasing robustness. This paper proposes a dynamic neural network framework based on data preprocessing and PDE integration to solve the above problems, which includes K-means Laplacian operator, information entropy, PDE integration, and dynamic network. The effectiveness of the proposed framework was evaluated on several distinct image datasets. Simulation results show that it has effective performance on databases, improving the initial and final accuracy of the network framework.