Embedding

An embedding in machine learning is a low-dimensional, continuous vector representation of a discrete, categorical variable, such as a word, user ID, or product item. Instead of using one-hot encodings which can be sparse and high-dimensional, embeddings transform these discrete entities into a dense vector space where similar items are mapped to nearby points. This transformation is not arbitrary; the position and relationships between vectors in the embedding space are learned during the training process to capture meaningful semantic or functional relationships between the original categorical variables.