What Is an Autoencoder?

Autoencoder Explainer

An autoencoder is a neural network trained to reproduce its input. An encoder maps the input into a latent representation, and a decoder uses that representation to reconstruct the original data. The difference between input and reconstruction supplies the training signal.

If the network can simply copy everything, it may learn an unhelpful identity function. Useful designs restrict the representation, add noise, encourage sparsity, or apply other regularization so that the model must capture important structure rather than memorize each example.

Autoencoders have been used for dimensionality reduction, denoising, anomaly detection, representation learning, and data compression. A low reconstruction error does not automatically mean the representation is meaningful for a particular downstream decision.

The decoder’s output reflects the model, loss, and training distribution. Unfamiliar inputs can reconstruct unpredictably, and anomaly thresholds require validation. Variational autoencoders add a probabilistic latent model, but a basic autoencoder need not be generative in that sense.

An autoencoder learns to compress an input into a representation and reconstruct the original input from that representation.

Key terms include encoder, decoder, latent representation, bottleneck, reconstruction, reconstruction loss, denoising, sparsity and anomaly detection.

A smaller, noisy, or regularized latent representation limits direct copying and encourages the network to preserve recurring structure.

The decoder then tests whether that representation contains enough information to reconstruct the input.

A model may reconstruct common patterns well while erasing rare but important details or reproducing anomalies it has learned.

Downstream usefulness and anomaly thresholds must therefore be evaluated on realistic held-out data.

Explore more "Explainers"

Discover additional explainers across politics, science, business, technology, and other fields. Each explainer breaks down a complex idea into clear, everyday language—helping you better understand how major concepts, systems, and debates shape the world around us.