Evading network anomaly detection systems: formal reasoning and practical techniques

By Prahlad Fogla, Wenke Lee. 2006

197 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Networks and Communications, Safety, Robustness & Interpretability, Privacy & Security, Signal Processing

How researchers see this paper

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

In Artificial Intelligence

In Computer Networks and Communications

In Safety, Robustness & Interpretability

In Privacy & Security

In Signal Processing

Citations in the corpus

The Limitations of Deep Learning in Adversarial Settings (2015, background): “We believe this makes adversarial crafting much easier for input domains like malware executables, which are not as easy to perturb as images [10], [15].”
Practical Evasion of a Learning-Based Classifier: A Case Study (2014, background): “Recent work has also witnessed several innovative applications of machine learning for detection of various kinds of security violations, e.g., drive-by-downloads [3], [4], malicious web pages [5], [6], compromised accounts and fake identities in social networks [7], [8], unwanted P2P traffic [9]…”
The security of machine learning (2010, background): “And finally, Fogla and Lee (2006) discuss in depth the opportunities and constraints of generating attack traffic that appears statistically identical to benign traffic, as mentioned in Sect.”
Security analysis of online centroid anomaly detection (2010, background): “In fact, finding a sequence with the best approximation to a given k-gram feature vector is NP-hard (Fogla and Lee, 2006).”
Machine learning in adversarial environments (2010, background): “Protection against adversarial data may seem to be a “mission impossible”.”
On the infeasibility of modeling polymorphic shellcode (2007, background): “• Spectrum shaping & byte padding: In polymorphic blending attacks [17], exploits attempt to appear similar t o benign traffic in terms of the n-gram content distribution.”
Practical Attacks Against Graph-based Clustering (2017, background): “For example, features are directly obtained from spam emails [36, 60], PDF files [49, 53, 62], phishing pages [49], images [16, 25, 44, 52], network attack packets [24], and exploits [55, 58].”
Adaptive Anomaly Detection via Self-calibration and Dynamic Updating (2009, background): “Perhaps more troubling, Fogla and Lee [8] have shown how to evade anomaly classifiers by constructing polymorphic exploits that blend with normal traffic (a sophisticated form of mimicry attack [28]), and Song et al.”
Open problems in the security of learning (2008, background): “intrusion detection systems [7, 27], and email spam filtering [16, 19, 30].”
Casting out Demons: Sanitizing Training Data for Anomaly Sensors (2008, background): “Although anomaly–based approaches are not perfect [9, 26, 27], recent research indicates that signature–based network intrusion detection systems are quickly becoming ineffective at identifying malicious traffic [4, 17, 23].”
When a Tree Falls: Using Diversity in Ensemble Classifiers to Identify Evasion in Malware Detectors (2016, methodology): “Some studies have proposed methods for creating effective classifier based intrusion detection systems [6], [15], [41].”
Machine Learning in the Presence of an Adversary: Attacking and Defending the SpamBayes Spam Filter (2008, background): “Fogla and Lee are able to fool intrusion detection systems through the use of a polymorphic blending attack [7], an attack that encrypts malicious traffic so as to appear identical to normal traffic.”
ANTIDOTE: understanding and defending against poisoning of anomaly detectors (2009, methodology): “Our third contribution is to carry out extensive evaluations of both antidote and the original PCA method, in a variety of poisoning situations, and to assess their performance via multiple metrics.”
Comprehensive Analysis and Detection of Flash-Based Malware (2016, methodology): “For n-gram models, Fogla and Lee [17] show that generating a polymorphic blending attack is NP-hard, but can be approximated for low-order n-grams.”
EagleEye: Attack-Agnostic Defense against Adversarial Inputs (Technical Report) (2018, background): “For example, crafting adversarial malware samples to evade malware detection may require adopting other metrics [13, 17].”
Outside the Closed World: On Using Machine Learning for Network Intrusion Detection (2010, background): “In [58], Fogla and Lee present an automated approach to mutate attacks so that they match a system’s normal profile.”