Latent Space
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
The compressed internal coordinate system a model learns, where each input becomes a vector and distance between points reflects similarity in meaning.
What is Latent Space?
A latent space is the compressed internal coordinate system a model learns, in which each input becomes a vector and distance between points reflects similarity. “Latent” means hidden: nobody designs the dimensions – the model invents them during training.
The space is far smaller than the raw input. A face photo is hundreds of thousands of pixel values; its latent representation might be a few hundred numbers that still capture what the face looks like.
How Latent Space Works
An encoder network maps each input to a point in the space. The bottleneck is the point: with so few numbers available, the model must keep what distinguishes one input from another and discard the rest.
What survives compression is structure. In a model trained on faces, inputs arrange themselves by pose, lighting, and age, and directions through the space correspond to meaningful variation even though no single coordinate is human-readable.
Word embeddings are a latent space for words. Generative models run the idea in reverse: pick a point in latent space and decode it into a new image, sound, or sentence.
Example of Latent Space
A variational autoencoder is trained on face photographs, compressing each image to a vector of, say, 200 numbers. Take two photos of different people and walk through what the space allows.
First, encode photo A and photo B into their two vectors. Second, average them, giving a point halfway between the two faces. Third, decode that midpoint: out comes a plausible new face blending both people.
Averaging raw pixels instead would produce a ghostly double exposure. The blend works only because the latent space is organized by similarity, so the midpoint of two faces still sits in face territory.
Related AI terms: Word Embedding · Embedding Layer · Autoencoder · Variational Autoencoder · Dimensionality Reduction
Did you like the Latent Space gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About