Dictionary of Applied Machine Learning
Updated on 2026-09-18
Typeset PDF version — the authoritative form of this entry
An autoencoder is a machine learning (ML) method that learns an encoder map $\hypothesis: \featurevec \mapsto \vz$ and a decoder map $\hypothesis^{*}: \vz \mapsto \widehat{\featurevec}$ jointly, so that the reconstruction $\hypothesis^{*}(\hypothesis(\featurevec))$ is close to the original feature vector $\featurevec$. The code $\vz$ typically has fewer entries than $\featurevec$, which makes an autoencoder a dimensionality reduction and feature learning method. Training uses empirical risk minimization (ERM) with a loss that measures the reconstruction error, and no labels are needed, so it is a form of self-supervised learning. With linear models for encoder and decoder, the learned code space coincides with that of principal component analysis (PCA); artificial neural networks (ANNs) as encoder and decoder allow nonlinear codes.
An autoencoder is a machine learning (ML) method that simultaneously
learns an encoder map $\hypothesis \in \hypospace$ and a decoder map
$\hypothesis^{*} \in \hypospace^{*}$. Different autoencoders use different
models $\hypospace, \hypospace^{*}$, e.g., artificial neural networks (ANNs) with different architectures.
The special case of an autoencoder using (vector-valued) linear models for
$\hypospace, \hypospace^{*}$ results in principal component analysis (PCA).
See also: encoder, feature learning, dimensionality reduction.
@misc{dictml_autoencoder,
author = {Jung, Alexander and Olioumtsevits, Konstantina and Schnoor, Ekkehard},
title = {autoencoder},
howpublished = {Dictionary of Applied Machine Learning (course edition)},
year = {2026},
doi = {10.5281/zenodo.21569296},
note = {ISBN 978-952-64-3013-3, CC BY 4.0, retrieved 2026-09-19},
url = {https://dictionaryofml.org/terms/autoencoder.html}
}