What does an auto-encoder consist of?

An auto-encoder specifically consists of an encoder and a decoder. The encoder transforms the input data into a compressed, lower-dimensional representation, often referred to as the latent space. This compressed format captures the most essential features of the input. Following the encoding process, the decoder takes this representation and attempts to reconstruct the original input data as accurately as possible.

The primary purpose of an auto-encoder is to learn efficient representations of data, which can be useful for tasks such as dimensionality reduction or generating new data samples. The architecture can vary from a simple fully connected network to more complex structures like convolutional auto-encoders, depending on the nature of the input data (e.g., images, text). Overall, the encoder-decoder framework is fundamental to the operation of auto-encoders, making this choice the correct answer.