Is theoretical linear algebra at the level of axler helpful for machine learning? If so, in what ways?