What is dimensionality reduction in deep learning embeddings?

Embeddings reduce high-dimensional sparse inputs into dense low-dimensional vectors.

Neural networks learn embeddings for words, images, or users, compressing semantic information into dense vectors optimized during training.