A framework for quantitative security analysis of machine learning

We propose a framework for quantitative security analysis of machine learning methods. Key issus of this framework are a formal specification of the deployed learning model and an attackers constraints, the computation of an optimal attack, and a derivation of an upper bound on the adversarial impact. We exemplarily apply the framework for the analysis of one specific learning scenario, online centroid anomaly detection and experimentally verify the tightness of obtained theoretical bounds. Categories and Subject Descriptors C.2.0 [Computer-Communication Networks]: General—Security and protection; I.2.6 [Artificial Intelligence]: Learning—Parameter learning; I.5.2 [Pattern Recognition]: Design Methodology—Classifier design and evaluation General Terms Algorithms, Experimentation, Security Keywords Machine learning, computer security, centroid anomaly detection, intrusion detection, adversarial learning

• Type: Conference
• Year: 2009
• Where: CCS
• Authors: Pavel Laskov, Marius Kloft
• Keywords: CCS 2009, Centroid Anomaly Detection, Machine Learning, Machine Learning Methods, Security Privacy