Hands-On Artificial Intelligence for Beginners

Encoder

The encoder generates its distribution by first defining its prior as a standard normal distribution. Then, during training, this distribution becomes updated, and the decoder can easily sample from this distribution later on. Both the encoder and the decoder are unique in terms of VAEs in that they output two vectors instead of one: a vector of means, μ, and another vector of standard deviation, σ. These help to define the limits for our generated distributions. Intuitively, the mean vector controls where the encoding of an input should be centered, while the standard deviation controls the extent to which the encoding may vary from the mean. This constraint on the encoder forces the network to learn a distribution, thereby taking ...