Source: https://machine-learning-and-security.github.io/papers/mlsec17_paper_52.pdf