[uoaei]

All machine learning is just embedding of various forms. If you have a way to translate disparate types of data into a common space, in ways that preserve inductive bias and information content, you can then combine them for downstream tasks.