Adversarial Edit Attacks for Tree Data
This work addresses the vulnerability of tree classifiers in fields like medicine and automated program analysis, representing an incremental extension of adversarial attack methods to a new data type.
The paper tackles the problem of adversarial attacks on tree-structured data, which had been limited to vectorial data like images, by introducing adversarial edit attacks that use tree edit distance and a logarithmic number of black-box queries, and shows that many established tree classifiers can be effectively attacked on programming and biomedical datasets.
Many machine learning models can be attacked with adversarial examples, i.e. inputs close to correctly classified examples that are classified incorrectly. However, most research on adversarial attacks to date is limited to vectorial data, in particular image data. In this contribution, we extend the field by introducing adversarial edit attacks for tree-structured data with potential applications in medicine and automated program analysis. Our approach solely relies on the tree edit distance and a logarithmic number of black-box queries to the attacked classifier without any need for gradient information. We evaluate our approach on two programming and two biomedical data sets and show that many established tree classifiers, like tree-kernel-SVMs and recursive neural networks, can be attacked effectively.