Adversarial Machine Learning

By J. Doug Tygar. 2011

269 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Safety, Robustness & Interpretability

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 Safety, Robustness & Interpretability

Citations in the corpus

How To Backdoor Federated Learning (2018, background): “the model’s weights [60] or its accuracy on the main task. 2Related Work Training-time attacks.“Traditional” poisoning attacks compromise the training data to change the model’s behavior at test time [4, 30, 42, 58, 63]. Previous backdoor attacks change the model’s behavior only on specific attacker”
Distillation as a Defense to Adversarial Perturbations Against Deep Neural Networks (2015, background): “Such attacks commonly occur today in non-DL classification systems [12], [13], [14], [15], [16].”
Privacy Risks of Securing Machine Learning Models against Adversarial Examples (2019, neutral): “signed without considering potential adversarial threats, their security and privacy vulnerabilities have come to a forefront in recent years, together with the arms race between attacks and defenses [7, 22, 39]. In the security domain, the adversary aims to induce misclassifications to the target m”
Detecting Backdoor Attacks on Deep Neural Networks by Activation Clustering (2018, background): “ trained by third parties. In these scenarios, adversaries may be able to alter the model’s behavior by manipulating the data that is used to train it. Prior work (Barreno et al., 2010; Nelson, 2010; Huang et al., 2011; Papernot et al., 2016) has shown that this type of attack, called a poisoning at”
Practical Black-Box Attacks against Machine Learning (2016, methodology): “n inputs and a predeﬁned set of classes. For instance, a malware detector is a classiﬁer taking executables as inputs and assigning them either to the benign or malware class. Efforts in the security [1], [2], [3], [4] and machine learning [5], [6] communities exposed the vulnerability of machine le”
STRIP: a defence against trojan attacks on deep neural networks (2019, background): “Previous poisoning attacks usually aim to degrade a classifier’s accuracy of clean inputs [28], [29].”
Attacking Graph-based Classification via Manipulating the Graph Structure (2019, background): “ing studies on adversarial machine learning mainly focused on machine learning for non-graph data. In particular, studies have demonstrated that machine learning is vulnerable to adversarial examples [7,12,27,30,37,40,43,46,51,53,56, 68], poisoning attacks [8,18,31,38,67,70], privacy attacks for use”
When Does Machine Learning FAIL? Generalized Transferability for Evasion and Poisoning Attacks (2018, neutral): “Adversarial limitations are also discussed in [23].”
Mitigating Evasion Attacks to Deep Neural Networks via Region-based Classification (2017, background): “ension space into L class regions, denoted as R1, R2, , RL. Any data point in the region Ri will be predicted to have label i by the classi•er. 2.2 Evasion Attacks Poisoning a‹acks and evasion a‹acks [13] are two well-known attacks to machine learning/data mining. A poisoning a−ack aims to pollute t”
The Limitations of Deep Learning in Adversarial Settings (2015, background): “he security of machine learning [2] is an active research topic within the security and machine learning communities. A broad taxonomy of attacks and required adversarial capabilties are discussed in [22] and [3] along with considerations for building defense mechanisms. Biggio et al. studied classi”
Is Feature Selection Secure against Training Data Poisoning? (2015, methodology): “The worst-case scenario in which the attacker has full knowledge of the attacked system, is usually referred to as perfect knowledge case (Biggio et al., 2014; 2013b; 2012; Kloft & Laskov, 2010; Brückner et al., 2012; Huang et al., 2011; Barreno et al., 2006), and it allows one to empirically evalua”
Transferability in Machine Learning: from Phenomena to Black-Box Attacks using Adversarial Samples (2016, background): “mentation. This work is part of a series of security evaluations of machine learning algorithms [1, 5]. Unlike us, previous work in this eld assumed knowledge of the model architecture and parameters [6, 14]. Our threat model considered adversaries interested in misclassication at test time, once th”
Poisoning behavioral malware clustering (2014, methodology): “representation, as many other adversarial machine learning problems [5–7, 9, 17].”
Security Evaluation of Pattern Classifiers under Attack (2014, background): “extending the classical design cycle of [9]. In particular, three main open issues can be identiﬁed: (i) analyzing the vulnerabilities of classiﬁcation algorithms, and the corresponding attacks [11], [14], [15]; (ii) developing novel methods to assess classiﬁer security against these attacks, which ”
Is Deep Learning Safe for Robot Vision? Adversarial Examples Against the iCub Humanoid (2017, background): “As in previous work addressing the issue of evaluating security of machine-learning algorithms [1, 12, 5, 4, 28], our underlying idea is to evaluate the maximum recognition accuracy degradation against an increasing maximum admissible level of perturbation of the input images.”
Randomized Prediction Games for Adversarial Machine Learning (2016, background): “To date, several authors have addressed the problem of designing secure learning algorithms to mitigate the impact of evasion attacks [1], [6], [10], [11], [21]–[27] (see Section VII for further details).”
Adversarial Feature Selection Against Evasion Attacks (2016, methodology): “bustness to random perturbations, and vice versa. Finally, it is worth mentioning that Robust Statistics [62], [63] may be also exploited to learn more robust feature mappings. An example is given in [10], [26], where the authors have exploited a robust version of the principal component analysis (o”
Generative Poisoning Attack Method Against Neural Networks (2017, background): “art devices [3], and an audio sensing framework built from coupled DNNs [4]. Such applications, however, introduce at least two types of security issues – causative attack and exploratory attack [1], [8]. Exploratory attack does not change the parameters of the target model. Instead, the attacker se”
SoK: Security and Privacy in Machine Learning (2018, methodology): “Our development of the threat model in following sections builds on previous treatments [19], [20], [21].”