This article describes how the most widely used clustering, k-means, is prone to fall into a local optimum. Notably, traditional clustering approaches are directly performed on private data and fail to cope with malicious attacks in massive data mining tasks against attackers' arbitrary...