This paper presents a more efficient method for finding adversarial examples for tree ensembles -- that is, finding an instance with a different predicted label that is close to a given query instance. It's good to see work on adversarial attacks against tree ensembles, since they're widely used but rarely studied. This is a nice step forward.