Autoencoder
Also: AE, denoising autoencoder
An encoder compresses the input into a low-dimensional latent code and a decoder reconstructs the input from it, trained to minimize reconstruction error without labels. Used for compression, denoising and anomaly detection; a plain autoencoder is not generative.
L = ‖x − dec(enc(x))‖²
Example
An 8×8 pattern (64 values) is encoded to 2 numbers and decoded back to 64; inputs with high reconstruction error can be flagged as anomalies.