However, implementing trained mathematical transformations by designing hardware for strict, operation-by-operation mathematical isomorphism is not the only way to perform efficient machine learning.