Key points are not available for this paper at this time.
In machine learning, the concept of interpretability is both important and slippery.
Zachary C. Lipton (Fri,) studied this question.