Latent space

A latent space is a lower-dimensional representation of data that compresses complex, high-dimensional information into a simpler form while preserving the most important underlying factors. In this space, similar data points are positioned closer together, making patterns and relationships easier for machine learning models to detect and use.

Latent spaces are common in AI applications such as text or image generation, where data is encoded into embeddings, manipulated or sampled, and then decoded to produce new outputs.

To clarify the distinction, a feature space captures a broad set of observable or engineered characteristics of the data, whereas a latent space represents a compressed representation that retains only the factors most relevant to a given task.