Neural-Network-Based Feature Learning: Convolutional Neural Network

Abstract

Convolutional neural network (CNNs) are a kind of feedforward neural network with convolutional computation and deep structure. In recent years, the application of CNN is very extensive, such as visual images, video recognition, and natural language processing. In this chapter, we first introduce the basic architecture of CNN, including convolutional layers, pooling layers, batch normalization layers, and dropout layers, and pay more attention to the illustration of backpropagation of convolutional layers. Then we introduce transfer feature learning to use of similarities between data, tasks, or models to apply a model that has been learned in one field to a learning problem in another field. Later we introduce some popular and widely-used deep convolutional models, including AlexNet, VggNet, and GoogleNet. Finally, we use case studies to deepen the understanding of CNNs.