Chapter 3. Stacked Denoising Autoencoders

In this chapter, we'll continue building our skill with deep architectures by applying Stacked Denoising Autoencoders (SdA) to learn feature representations for high-dimensional input data.

We'll start, as before, by gaining a solid understanding of the theory and concepts that underpin autoencoders. We'll identify related techniques and call out the strengths of autoencoders as part of your data science toolkit. We'll discuss the use of Denoising Autoencoders (dA), a variation of the algorithm that introduces stochastic corruption to the input data, obliging the autoencoder to decorrupt the input and, in so doing, build a more effective feature representation.

We'll follow up on theory, as before, by walking through the code for a dA class, linking theory and implementation details to build a strong understanding of the technique.

At this point, we'll take a journey very similar to that taken in the preceding...