Chapter 3 - Deep learning

Abstract

The goal of developing artificial intelligence computing systems is to enable machines to better understand and serve humans. Machines understand vision and speech mainly based on deep learning. Deep learning refers to large-scale deep neural networks. In this chapter, we first introduce the basic convolutional neural network (CNN) [15] for image processing. Then, we introduce the principle and development of typical CNNs for image classification and object detection. Besides CNNs for image processing, we also introduce the recurrent neural network (RNN) [13] and its extension, long short-term memory (LSTM) [61], for sequential data including speech, text, and video. Next, we will discuss a relatively novel deep learning technology, generative adversarial networks [71], which has many interesting applications. Finally, this chapter ends with an implementation of the driving example—image style transfer.