Strengthening Deep Neural Networks

Chapter 4. DNN Processing for Image, Audio, and Video

In Chapter 3, we created a neural network able to classify simple images of items of clothing to illustrate the principles of neural network technology. However, the simplicity of the Fashion-MNIST data meant that it was not a scenario with much realism. DNNs are able to solve far more complex problems through a wide range of architectures other than simple feed-forward, fully connected networks.

There are many different ways of designing a DNN, using different layering structures, different types of layers, and different ways of connecting the nodes. The network will depend upon the task, and these different flavors of DNN may also be combined with each other or other algorithms, enabling them to perform even more complex tasks.

This chapter, therefore, considers how DNN technology can be applied to more realistic image, audio, and video processing. We’ll build on the basics from Chapter 3 to understand the concepts behind the neural networks commonly used in these fields and how these technologies are combined with more traditional processing to get the best results.

This discussion will focus on two key network types: convolutional neural networks (CNNs) and recurrent neural networks (RNNs). These demonstrate common patterns in the context of image and audio processing and also are referenced in discussions later in the book. CNNs and RNNs are not the only DNNs relevant to image, audio, and video processing, but they are ...