Poisoning Complete-Linkage Hierarchical Clustering

摘要: Clustering algorithms are largely adopted in security applications as a vehicle to detect malicious activities, although few attention has been paid on preventing deliberate attacks from subverting the clustering process itself. Recent work introduced methodology for analysis of data adversarial settings, aimed identify potential against and evaluate their impact. The authors have shown that single-linkage hierarchical can be severely affected by presence very small fraction carefully-crafted poisoning into input data, highlighting algorithm may itself weakest link system. In this paper, we extend case complete-linkage devising an ad hoc attack. We verify its effectiveness artificial application examples related malware handwritten digits.