Compromising PCA-based Anomaly Detectors for Network-Wide Traffic

33 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Using Machine Teaching to Identify Optimal Training-Set Attacks on Machine Learners (2015, background): “Introduction The study on security threats to intelligent agents has a long history (Nelson et al. 2009; Tan, Killourhy, and Maxion 2002; Rubinstein et al. 2008; Barreno et al. 2010; Laskov and Lippmann 2010; Barreno et al. 2006; Dalvi et al. 2004; Liu and Chawla 2009; Biggio, Fumera, and Roli 2013;”
Behavior of Machine Learning Algorithms in Adversarial Environments (2010, methodology): “For each scenario, I present only one representative poisoning scheme, although others were studied in prior work [Rubinstein et al., 2008].”
Open problems in the security of learning (2008, methodology): “[25] outline data poisoning attacks for increasing false negatives in principal components analysis (PCA), as used in network-wide volume anomaly detection by Lakhina et al.”
Stealthy poisoning attacks on PCA-based anomaly detectors (2009, background): “We previously showed that an adversary can generate OD traffic flow patterns that mislead this network anomography technique and lead it to miss anomalous traffic flows [5, 6].”
Towards Detecting Anomalous User Behavior in Online Social Networks (2014, background): “Compromising Training: An attacker that controls a sufficiently large number of users may attempt to compromise the training phase by injecting additional likes, thereby distorting the principal components learned for normal users [39, 55, 56].”