It's branding (see: TensorFlow); also, pretty much anything (linear) you would do with an arbitrarily ranked tensor can be expressed in terms of vector ops and matmuls