Biblio

Machine learning (ML) models are increasingly being used in the development of Malware Detection Systems. Existing research in this area primarily focuses on developing new architectures and feature representation techniques to improve the accuracy of the model. However, recent studies have shown that existing state-of-the art techniques are vulnerable to adversarial machine learning (AML) attacks. Among those, data poisoning attacks have been identified as a top concern for ML practitioners. A recent study on clean-label poisoning attacks in which an adversary intentionally crafts training samples in order for the model to learn a backdoor watermark was shown to degrade the performance of state-of-the-art classifiers. Defenses against such poisoning attacks have been largely under-explored. We investigate a recently proposed clean-label poisoning attack and leverage an ensemble-based Nested Training technique to remove most of the poisoned samples from a poisoned training dataset. Our technique leverages the relatively large sensitivity of poisoned samples to feature noise that disproportionately affects the accuracy of a backdoored model. In particular, we show that for two state-of-the art architectures trained on the EMBER dataset affected by the clean-label attack, the Nested Training approach improves the accuracy of backdoor malware samples from 3.42% to 93.2%. We also show that samples produced by the clean-label attack often successfully evade malware classification even when the classifier is not poisoned during training. However, even in such scenarios, our Nested Training technique can mitigate the effect of such clean-label-based evasion attacks by recovering the model's accuracy of malware detection from 3.57% to 93.2%.

5G and beyond 5G low power wireless networks make Internet of Things (IoT) and Cyber-Physical Systems (CPS) applications capable of serving massive amounts of devices and machines. Due to the broadcast nature of wireless networks, it is crucial to secure the communication between these devices and machines from spoofing and interception attacks. This paper is concerned with the security of carrier frequency offset (CFO) based continuous physical layer authentication. The interaction between an attacker and a defender is modeled as a dynamic discrete leader-follower game with imperfect information. In the considered model, a legitimate user (Alice) communicates with the defender/operator (Bob) and is authorized by her CFO continuously. The attacker (Eve), by listening/eavesdropping the communication between Alice and Bob, tries to learn the CFO characteristics of Alice and aims to inject malicious packets to Bob by impersonating Alice. First, by showing that the optimal attacker strategy is a threshold policy, an optimization problem of the attacker with exponentially growing action space is reduced to a tractable integer optimization problem with a single parameter, then the corresponding defender cost is derived. Extensive simulations illustrate the characteristics of optimal strategies/utilities of the players depending on the actions, and show that the defender’s optimal false positive rate causes attack success probabilities to be in the order of 0.99. The results show the importance of the parameters while finding the balance between system security and efficiency.

Recently deep learning has demonstrated much success within the fields of image and natural language processing, facial recognition, and computer vision. The success is attributed to large, accessible databases and deep learning's ability to learn highly accurate models. Thus, deep learning is being investigated as a viable end-to-end approach to digital communications design. This work investigates the use of adversarial deep learning to ensure that a radio can communicate covertly, via Direct Sequence Spread Spectrum (DSSS), with another while a third (the adversary) is actively attempting to detect, intercept and exploit their communications. The adversary's ability to detect and exploit the DSSS signals is hindered by: (i) generating a set of spreading codes that are balanced and result in low side lobes as well as (ii) actively adapting the encoding scheme. Lastly, DSSS communications performance is assessed using energy constrained devices to accurately portray IoT and IoBT device limitations.

Whilst machine learning has been widely adopted for various domains, it is important to consider how such techniques may be susceptible to malicious users through adversarial attacks. Given a trained classifier, a malicious attack may attempt to craft a data observation whereby the data features purposefully trigger the classifier to yield incorrect responses. This has been observed in various image classification tasks, including falsifying road sign detection and facial recognition, which could have severe consequences in real-world deployment. In this work, we investigate how these attacks could impact on network traffic analysis, and how a system could perform misclassification of common network attacks such as DDoS attacks. Using the CICIDS2017 data, we examine how vulnerable the data features used for intrusion detection are to perturbation attacks using FGSM adversarial examples. As a result, our method provides a defensive approach for assessing feature robustness that seeks to balance between classification accuracy whilst minimising the attack surface of the feature space.

Adversarial machine learning calculations handle adversarial instance age, producing bogus data information with the ability to fool any machine learning model. As the word implies, “foe” refers to a rival, whereas “rival” refers to a foe. In order to strengthen the machine learning models, this section discusses about the weakness of machine learning models and how effectively the misinterpretation occurs during the learning cycle. As definite as it is, existing methods such as creating adversarial models and devising powerful ML computations, frequently ignore semantics and the general skeleton including ML section. This research work develops an adversarial learning calculation by considering the coordinated portrayal by considering all the characteristics and Convolutional Neural Networks (CNN) explicitly. Figuring will most likely express minimal adjustments via data transport represented over positive and negative class markings, as well as a specific subsequent data flow misclassified by CNN. The final results recommend a certain game theory and formative figuring, which obtain incredible favored ensuring about significant learning models against the execution of shortcomings, which are reproduced as attack circumstances against various adversaries.

In the past decades, learning an effective distance metric between pairs of instances has played an important role in the classification and retrieval task, for example, the person identification or malware retrieval in the IoT service. The core motivation of recent efforts focus on improving the metric forms, and already showed promising results on the various applications. However, such models often fail to produce a reliable metric on the ambiguous test set. It happens mainly due to the sampling process of the training set, which is not representative of the distribution of the negative samples, especially the examples that are closer to the boundary of different categories (also called hard negative samples). In this paper, we focus on addressing such problems and propose an adaptive margin deep adversarial metric learning (AMDAML) framework. It exploits numerous common negative samples to generate potential hard (adversarial) negatives and applies them to facilitate robust metric learning. Apart from the previous approaches that typically depend on the search or data augmentation to find hard negative samples, the generation of adversarial negative instances could avoid the limitation of domain knowledge and constraint pairs' amount. Specifically, in order to prevent over fitting or underfitting during the training step, we propose an adaptive margin loss that preserves a flexible margin between the negative (include the adversarial and original) and positive samples. We simultaneously train both the adversarial negative generator and conventional metric objective in an adversarial manner and learn the feature representations that are more precise and robust. The experimental results on practical data sets clearly demonstrate the superiority of AMDAML to representative state-of-the-art metric learning models.

The regularity of devastating cyber-attacks has made cybersecurity a grand societal challenge. Many cybersecurity professionals are closely examining the international Dark Web to proactively pinpoint potential cyber threats. Despite its potential, the Dark Web contains hundreds of thousands of non-English posts. While machine translation is the prevailing approach to process non-English text, applying MT on hacker forum text results in mistranslations. In this study, we draw upon Long-Short Term Memory (LSTM), Cross-Lingual Knowledge Transfer (CLKT), and Generative Adversarial Networks (GANs) principles to design a novel Adversarial CLKT (A-CLKT) approach. A-CLKT operates on untranslated text to retain the original semantics of the language and leverages the collective knowledge about cyber threats across languages to create a language invariant representation without any manual feature engineering or external resources. Three experiments demonstrate how A-CLKT outperforms state-of-the-art machine learning, deep learning, and CLKT algorithms in identifying cyber-threats in French and Russian forums.

IoT malware detection using control flow graph (CFG)-based features and deep learning networks are widely explored. The main goal of this study is to investigate the robustness of such models against adversarial learning. We designed two approaches to craft adversarial IoT software: off-the-shelf methods and Graph Embedding and Augmentation (GEA) method. In the off-the-shelf adversarial learning attack methods, we examine eight different adversarial learning methods to force the model to misclassification. The GEA approach aims to preserve the functionality and practicality of the generated adversarial sample through a careful embedding of a benign sample to a malicious one. Intensive experiments are conducted to evaluate the performance of the proposed method, showing that off-the-shelf adversarial attack methods are able to achieve a misclassification rate of 100%. In addition, we observed that the GEA approach is able to misclassify all IoT malware samples as benign. The findings of this work highlight the essential need for more robust detection tools against adversarial learning, including features that are not easy to manipulate, unlike CFG-based features. The implications of the study are quite broad, since the approach challenged in this work is widely used for other applications using graphs.

Currently, fusing visible and thermal images for facial expression recognition requires two modalities during both training and testing. Visible cameras are commonly used in real-life applications, and thermal cameras are typically only available in lab situations due to their high price. Thermal imaging for facial expression recognition is not frequently used in real-world situations. To address this, we propose a novel thermally enhanced facial expression recognition method which uses thermal images as privileged information to construct better visible feature representation and improved classifiers by incorporating adversarial learning and similarity constraints during training. Specifically, we train two deep neural networks from visible images and thermal images. We impose adversarial loss to enforce statistical similarity between the learned representations of two modalities, and a similarity constraint to regulate the mapping functions from visible and thermal representation to expressions. Thus, thermal images are leveraged to simultaneously improve visible feature representation and classification during training. To mimic real-world scenarios, only visible images are available during testing. We further extend the proposed expression recognition method for partially unpaired data to explore thermal images' supplementary role in visible facial expression recognition when visible images and thermal images are not synchronously recorded. Experimental results on the MAHNOB Laughter database demonstrate that our proposed method can effectively regularize visible representation and expression classifiers with the help of thermal images, achieving state-of-the-art recognition performance.

Recently researchers have proposed using deep learning-based systems for malware detection. Unfortunately, all deep learning classification systems are vulnerable to adversarial learning-based attacks, or adversarial attacks, where miscreants can avoid detection by the classification algorithm with very few perturbations of the input data. Previous work has studied adversarial attacks against static analysis-based malware classifiers which only classify the content of the unknown file without execution. However, since the majority of malware is either packed or encrypted, malware classification based on static analysis often fails to detect these types of files. To overcome this limitation, anti-malware companies typically perform dynamic analysis by emulating each file in the anti-malware engine or performing in-depth scanning in a virtual machine. These strategies allow the analysis of the malware after unpacking or decryption. In this work, we study different strategies of crafting adversarial samples for dynamic analysis. These strategies operate on sparse, binary inputs in contrast to continuous inputs such as pixels in images. We then study the effects of two, previously proposed defensive mechanisms against crafted adversarial samples including the distillation and ensemble defenses. We also propose and evaluate the weight decay defense. Experiments show that with these three defenses, the number of successfully crafted adversarial samples is reduced compared to an unprotected baseline system. In particular, the ensemble defense is the most resilient to adversarial attacks. Importantly, none of the defenses significantly reduce the classification accuracy for detecting malware. Finally, we show that while adding additional hidden layers to neural models does not significantly improve the malware classification accuracy, it does significantly increase the classifier's robustness to adversarial attacks.

Unlike traditional learning to rank models that depend on hand-crafted features, neural representation learning models learn higher level features for the ranking task by training on large datasets. Their ability to learn new features directly from the data, however, may come at a price. Without any special supervision, these models learn relationships that may hold only in the domain from which the training data is sampled, and generalize poorly to domains not observed during training. We study the effectiveness of adversarial learning as a cross domain regularizer in the context of the ranking task. We use an adversarial discriminator and train our neural ranking model on a small set of domains. The discriminator provides a negative feedback signal to discourage the model from learning domain specific representations. Our experiments show consistently better performance on held out domains in the presence of the adversarial discriminator–-sometimes up to 30% on precision\$@1\$.

Sentiment analysis and other practices for text analytics on social media rely on publicly available and editable collections of data for training and evaluation. These data collections are subject to poisoning and data contamination attacks by adversaries having an interest in misleading the results of the performed analysis. We present the problem of adversarial text mining with a focus on decision making and we suggest cross-discipline, cross-application and cross-model strategies for more robust analyses. Our approach is practitioner-centric and is based on broadly-used interpretable models with applications in decision making.

Machine learning (ML) is commonly used in multiple disciplines and real-world applications, such as information retrieval, financial systems, health, biometrics and online social networks. However, their security profiles against deliberate attacks have not often been considered. Sophisticated adversaries can exploit specific vulnerabilities exposed by classical ML algorithms to deceive intelligent systems. It is emerging to perform a thorough security evaluation as well as potential attacks against the machine learning techniques before developing novel methods to guarantee that machine learning can be securely applied in adversarial setting. In this paper, an effective attack strategy for crafting foreign support vectors in order to attack a classic ML algorithm, the Support Vector Machine (SVM) has been proposed with mathematical proof. The new attack can minimize the margin around the decision boundary and maximize the hinge loss simultaneously. We evaluate the new attack in different real-world applications including social spam detection, Internet traffic classification and image recognition. Experimental results highlight that the security of classifiers can be worsened by poisoning a small group of support vectors.

Poisoning attack in which an adversary misleads the learning process by manipulating its training set significantly affect the performance of classifiers in security applications. This paper proposed a robust learning method which reduces the influences of attack samples on learning. The sensitivity, defined as the fluctuation of the output with small perturbation of the input, in Localized Generalization Error Model (L-GEM) is measured for each training sample. The classifier's output on attack samples may be sensitive and inaccurate since these samples are different from other untainted samples. An import score is assigned to each sample according to its localized generalization error bound. The classifier is trained using a new training set obtained by resampling the samples according to their importance scores. RBFNN is applied as the classifier in experimental evaluation. The proposed model outperforms than the traditional one under the well-known label flip poisoning attacks including nearest-first and farthest-first flips attack.

Deep neural networks (DNNs) are one of the most prominent technologies of our time, as they achieve state-of-the-art performance in many machine learning tasks, including but not limited to image classification, text mining, and speech processing. However, recent research on DNNs has indicated ever-increasing concern on the robustness to adversarial examples, especially for security-critical tasks such as traffic sign identification for autonomous driving. Studies have unveiled the vulnerability of a well-trained DNN by demonstrating the ability of generating barely noticeable (to both human and machines) adversarial images that lead to misclassification. Furthermore, researchers have shown that these adversarial images are highly transferable by simply training and attacking a substitute model built upon the target model, known as a black-box attack to DNNs. Similar to the setting of training substitute models, in this paper we propose an effective black-box attack that also only has access to the input (images) and the output (confidence scores) of a targeted DNN. However, different from leveraging attack transferability from substitute models, we propose zeroth order optimization (ZOO) based attacks to directly estimate the gradients of the targeted DNN for generating adversarial examples. We use zeroth order stochastic coordinate descent along with dimension reduction, hierarchical attack and importance sampling techniques to efficiently attack black-box models. By exploiting zeroth order optimization, improved attacks to the targeted DNN can be accomplished, sparing the need for training substitute models and avoiding the loss in attack transferability. Experimental results on MNIST, CIFAR10 and ImageNet show that the proposed ZOO attack is as effective as the state-of-the-art white-box attack (e.g., Carlini and Wagner's attack) and significantly outperforms existing black-box attacks via substitute models.

Following the recent adoption of deep neural networks (DNN) accross a wide range of applications, adversarial attacks against these models have proven to be an indisputable threat. Adversarial samples are crafted with a deliberate intention of undermining a system. In the case of DNNs, the lack of better understanding of their working has prevented the development of efficient defenses. In this paper, we propose a new defense method based on practical observations which is easy to integrate into models and performs better than state-of-the-art defenses. Our proposed solution is meant to reinforce the structure of a DNN, making its prediction more stable and less likely to be fooled by adversarial samples. We conduct an extensive experimental study proving the efficiency of our method against multiple attacks, comparing it to numerous defenses, both in white-box and black-box setups. Additionally, the implementation of our method brings almost no overhead to the training procedure, while maintaining the prediction performance of the original model on clean samples.