Proving Data-Poisoning Robustness in Decision Trees
We present a sound verification technique based on abstract interpretation and implement it in a tool called Antidote, which abstractly trains decision trees for an intractably large space of possible poisoned datasets.
