Hands-On Deep Learning for IoT

Autoencoders

AEs are also special types of neural networks that learn automatically from the input data. AEs consist of two components: the encoder and the decoder. The encoder compresses the input into a latent-space representation. Then, the decoder part tries to reconstruct the original input data from this representation:

• Encoder: Encodes or compresses the input into a latent-space representation using a function known as h = f(x)
• Decoder: Decodes or reconstructs the input from the latent space representation using a function known as r = g(h)

So, an AE can be described by a function of g(f(x)) = 0, where we want 0 as close to the original input of x. The following diagram shows how an AE typically works:

AEs are very useful for data ...