Keras documentation isn\'t clear what this actually is. I understand we can use this to compress the input feature space into a smaller one. But how is this done from a neur
In simple words (from the functionality point of view), it is a one-hot encoder and fully-connected layer. The layer weights are trainable.