Introduction

In the previous chapters, different network architectures were explained – from traditional ANNs, which can solve both classification and regression problems, to CNNs, which are mainly used to solve computer vision problems by performing the tasks of object classification, localization, detection, and segmentation.

In this final chapter, we will explore the concept of RNNs and solve sequential data problems. These network architectures are capable of handling sequential data where context is crucial, thanks to their ability to hold information from previous predictions, which is called memory. This means that, for instance, when analyzing a sentence word by word, RNNs have the ability to hold information about the first word of the sentence when they are handling the last one.

This chapter will explore the LSTM network architecture, which is a type of RNN that can hold both long-term and short-term memory and is especially useful for handling long sequences...