Ant Hole: Data Poisoning Attack Breaking out the Boundary of Face Cluster

With the continuous improvement of the open ability of machine learning, more and more users are benefited. Therefore, the impact on its security also expands. However, most of the research on the security of machine learning focuses on supervised learning, while the security of unsupervised learning has not been paid enough attention. In this paper, we propose a data poisoning attack method aimed at the face clustering open source project. We innovatively propose a fusion iterative method. It can smoothly generate a series of fusion face images which will fool the clustering algorithm.