Machine Learning

Latent Space

A compressed, lower-dimensional representation of data learned by a model. Points in latent space capture the essential features of the data, and nearby points represent similar data items.

Why It Matters

Latent spaces are how generative models create new content — they learn a meaningful map of possibilities and can sample new points to generate novel outputs.

Example

In a face-generating model, one direction in latent space might control age, another controls hair color, and moving along these axes generates predictable variations.

Think of it like...

Like a mixing board in a recording studio where each slider controls a different aspect of sound — the combination of slider positions defines a unique output.

Related Terms