Intermediate · Neural Networks

Latent Space

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. A compressed, lower-dimensional representation of data discovered by an unsupervised learning model.

Technical Definition

A compressed, lower-dimensional representation of data discovered by an unsupervised learning model.

How it works

Also known as a 'feature space' or 'embedding space,' the latent space captures the underlying structure and meaningful features of the input data. Elements in this space are typically vectors representing complex data points (like images or text) in a simplified, continuous form. Models like autoencoders or generative adversarial networks learn to compress and reconstruct data through this space, enabling tasks such as anomaly detection, data generation, and semantic understanding.

Related Concepts

• Embedding — A dense vector representation that captures semantic meaning, mapping discrete items like words into continuous mathematical space.
• Autoencoder — A neural network that learns compressed representations by training to reconstruct its own input through a bottleneck layer.