Deep Learning

18 Autoencoders

This chapter is about a particular kind of learning architecture called an autoencoder. One way to think about a standard autoencoder is that it’s a mechanism for compressing input, so it takes up less disk space and can be communicated more quickly, much as an MP3 encoder compresses music, or a JPG encoder compresses an image. The autoencoder gets its name from the idea that it automatically learns, by virtue of training, how best to encode, or represent, the input data. In practice, we usually use autoencoders for two types of jobs: removing the noise from a dataset, and reducing the dimensionality of a dataset.

We begin ...