Artificial Intelligence By Example

The use of autoencoders

The use of autoencoders must be explored although not recommended for the generation of a conceptual learning dataset.

Data compression has become a part of our daily life through MP3 algorithms, for example. Autoencoding in machine learning is a data compression algorithm.

It takes the original data, compresses it, and can then reconstruct it:

• Original input
• The encoder compresses the data (compressed representation)
• The decoder reconstructs the data
• The reconstructed output is produced

They were used for CNNs at a time for data representations. But as shown in Chapter 9, Getting Your Neurons to Work, random weight initialization scenarios are as effective if not more reliable and flexible.

Two main limitations ...