Limits of Learning-based Signature Generation with Adversaries

By Shobha Venkataraman, Avrim Blum, Dawn Song. 2008

63 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.

Citations in the corpus

Yes, Machine Learning Can Be More Secure! A Case Study on Android Malware Detection (2017, neutral): “quently cause misclassification errors [7], [11], [25], [32], [34], [42], [43], [44], [45].”
A Survey on Malware Detection Using Data Mining Techniques (2017, methodology): “— Adversarial Learning: In malware detection, using the data mining approaches opens the possibility for the malware attachers to come up with ways to “mistrain” the classiﬁers (e.g., by changing the data distribution or feature importance) [Dalvi et al. 2004; Venkataraman et al. 2008].”
Open problems in the security of learning (2008, background): “present lower bounds on the number of mistakes any supervised classifier must make when generating worm signatures in the presence of an adversary with full information and control [29].”
Adversarial machine learning (2019, background): “also present lower bounds for learning worm signatures based on red herring attacks [63].”
ANTIDOTE: understanding and defending against poisoning of anomaly detectors (2009, methodology): “Instead of modeling the squared prediction error as Gaussian (as in the original PCA method), we model the error using a Laplace distribution.”
All You Ever Wanted to Know about Dynamic Taint Analysis and Forward Symbolic Execution (but Might Have Been Afraid to Ask) (2010, unknown): “Due to the high overhead of binary instrumentation techniques, more efficient compiler-based [41, 63] and hardware-based [24, 25, 58, 59] approaches were later proposed.”
Man vs. Machine: Practical Adversarial Detection of Malicious Crowdsourcing Workers (2014, background): “Several studies have examined attacks on specific MLbased applications, from email spam detection [12] to network intuition detection [37, 40, 44] to malicious (PDF) file classification [5, 25, 41] and malware detection [21].”
A framework for quantitative security analysis of machine learning (2009, background): “In view of the recent surge of interest in machine learning applications in security – accompanied, however, by some skepticism whether machine learning can be ever made secure [1, 10, 12] – such reasoning may be instrumental in steering research into the right direction.”
Adversarial attacks against intrusion detection systems: Taxonomy, solutions and open issues (2013, background): “A purely manual process in the definition of classification models may be too expensive and too slow to cope with current threats, characterized by complex and rapidly evolving environmental settings [22], [158].”
Automatic analysis of malware behavior using machine learning (2011, unknown): “Hence, the most promising evasion attack against our framework corresponds to triggering fake and real malicious behavior during individual executions, similar to evasion techniques against signature generation [44,67].”
Behavior of Machine Learning Algorithms in Adversarial Environments (2010, methodology): “Independently, Wagner and Soto [2002] have also developed mimicry attacks against a sequence-based IDS called pH proposed by Somayaji and Forrest [2000].”
Traffic Morphing: An Efficient Defense Against Statistical Traffic Analysis (2009, background): “Finally, in Section 3.5 we discuss several potential issues that may arise in practice, and methods to overcome them.”