Embedding in machine learning refers to a representation learning technique that maps complex, high-dimensional data into a lower-dimensional vector space of numerical vectors.