autoencoder

This term is a technical compound used primarily in machine learning and data science. It describes a specific architecture where the network is trained to copy its input to its output, forcing the system to capture the most salient features of the data in a bottleneck layer.

In professional discourse, it is often distinguished from a standard encoder by the presence of a decoder component. While it can be used as a standalone tool for dimensionality reduction, it is frequently employed as a pre-training step for more complex deep learning models to initialize weights effectively.