Embedding space Machine Learning
noun phrase
Definition: A vector space into which data points, features, or symbols are mapped as learned embeddings so that geometrical relations in the space encode task-relevant structure, such as semantic, visual, or functional similarity. In many machine-learning contexts, similar items tend to be located relatively close to one another in the embedding space. Latent space is sometimes used as a near-equivalent term, but it is not always a strict synonym: embedding space usually refers to the learned representational space itself, whereas latent space may denote a broader class of hidden representational spaces, including those not explicitly used as embeddings. [Google ML Glossary].
Example in context: “Qualitative analysis suggests that the model aligns semantically related audio and visual features to particular dimensions of the embedding space.” [Chrupała 2022]
Synonym: latent space
Related terms: latent feature space, representation space