Visualization of Kernel Function of Convolutional Neural Network
Chengfan Li, Zirong Hu, Xuehai Ding · Journal of Physics Conference Series · 2020
Abstract In recent years, deep learning algorithms have been applied in various fields, and with the continuous development of neural networks, the newly emerging network structures have become more and more complex. However, neural networks are often similar to a black box. The evolution of parameters and the changes of neurons are unknowable. How to understand neural networks is an important research topic. Therefore, for beginners and researchers, a correct way to understand and explain neural networks is needed to improve neural networks. Visualization methods have always been used in the interpretability research of knowledge, which is very helpful to show complex algorithms and networks in the form of specific images, so that users and researchers have significant help in studying the structure of the algorithm and optimizing the performance of the algorithm. The visualization method has been applied to many mining fields. The maximization activation function was been used to realize the visualization of the kernel function in the convolutional neural network (CNN) in this work, and then the network structure and other content are visualized.