Vectors, dot products & layers

The real substrate of deep learning: a vector describes something, a dot product scores how well two descriptions match, and a layer runs many pattern-matches at once.