Introduction

This chapter continues our discussion of dimensionality reduction techniques as we turn our attention to autoencoders. Autoencoders are a particularly interesting area of focus as they provide a means of using supervised learning based on artificial neural networks, but in an unsupervised context. Being based on artificial neural networks, autoencoders are an extremely effective means of dimensionality reduction, but also provide additional benefits. With recent increases in the availability of data, processing power, and network connectivity, autoencoders are experiencing a resurgence in usage and study from their origins in the late 1980s. This is also consistent with the study of artificial neural networks, which was first described and implemented as a concept in the 1960s. Presently, you would only need to conduct a cursory internet search to discover the popularity and power of neural nets.

Autoencoders can be used for de-noising images and generating artificial data samples...