What Learned Representations and Influence Functions Can Tell Us About Adversarial Examples F i n d i n g s - I J C N L P