Search

Existing literature on adversarial Machine Learning (ML) focuses either on showing attacks that break every ML model or defenses that withstand most attacks. Unfortunately, little consideration is given to the actual feasibility of the attack or the defense. Moreover, adversarial samples are often crafted in the "feature-space," making the corresponding evaluations of questionable value. Simply put, the current situation does not allow one to estimate the actual threat posed by adversarial attacks, leading to a lack of secure ML systems. We aim to clarify such confusion in this article. By considering the application of ML for Phishing Website Detection (PWD), we formalize the "evasion-space," in which an adversarial perturbation can be introduced to fool an ML-PWD—demonstrating that even perturbations in the "feature-space" are useful. Then, we propose a realistic threat model describing evasion attacks against ML-PWD that are cheap to stage, and hence intrinsically more attractive for real phishers. After that, we perform the first statistically validated assessment of state-of-the-art ML-PWD against 12 evasion attacks. Our evaluation shows (i) the true efficacy of evasion attempts that are more likely to occur; and (ii) the impact of perturbations crafted in different evasion-spaces; our realistic evasion attempts induce a statistically significant degradation (3–10% at p< 0.05), and their cheap cost makes them a subtle threat. Notably, however, some ML-PWD are immune to our most realistic attacks (p = 0.22). Finally, as an additional contribution of this journal publication, we are the first to propose and empirically evaluate the intriguing case wherein an attacker introduces perturbations in multiple evasion-spaces at the same time. These new results show that simultaneously applying perturbations in the problem- and feature-space can cause a drop in the detection rate from 0.95 to 0. Our contribution paves the way for a much-needed re-assessment of adversarial attacks against ML systems for cybersecurity. [ABSTRACT FROM AUTHOR]

An introduction is presented in which the editor discusses articles in the issue on topics including the integrity of multimedia and multimodal data in the Internet of Things (IoT), privacy protection methods, and advanced data analysis techniques.

Video surveillance systems provide means to detect the presence of potentially malicious drones in the surroundings of critical infrastructures. In particular, these systems collect images and feed them to a deep-learning classifier able to detect the presence of a drone in the input image. However, current classifiers are not efficient in identifying drones that disguise themselves with the image background, e.g., hiding in front of a tree. Furthermore, video-based detection systems heavily rely on the image's brightness, where darkness imposes significant challenges in detecting drones. Both these phenomena increase the possibilities for attackers to get close to critical infrastructures without being spotted and hence be able to gather sensitive information or cause physical damages, possibly leading to safety threats. In this article, we propose RANGO, a drone detection arithmetic able to detect drones in challenging images where the target is difficult to distinguish from the background. RANGO is based on a deep learning architecture that exploits a Preconditioning Operation (PREP) that highlights the target by the difference between the target gradient and the background gradient. The idea is to highlight features that will be useful for classification. After PREP, RANGO uses multiple convolution kernels to make the final decision on the presence of the drone. We test RANGO on a drone image dataset composed of multiple already-existing datasets to which we add samples of birds and planes. We then compare RANGO with multiple currently existing approaches to show its superiority. When tested on images with disguising drones, RANGO attains an increase of 6.6% mean Average Precision (mAP) compared to YOLOv5 solution. When tested on the conventional dataset, RANGO improves the mAP by approximately 2.2%, thus confirming its effectiveness also in the general scenario. [ABSTRACT FROM AUTHOR]

Review platforms are viral online services where users share and read opinions about products (e.g., a smartphone) or experiences (e.g., a meal at a restaurant). Other users may be influenced by such opinions when deciding what to buy. The usability of review platforms is currently limited by the massive number of opinions on many products. Therefore, showing only the most helpful reviews for each product is in the best interest of both users and the platform (e.g., Amazon). The current state of the art is far from accurate in predicting how helpful a review is. First, most existing works lack compelling comparisons as many studies are conducted on datasets that are not publicly available. As a consequence, new studies are not always built on top of prior baselines. Second, most existing research focuses only on features derived from the review text, ignoring other fundamental aspects of the review platforms (e.g., the other reviews of a product, the order in which they were submitted). In this article, we first carefully review the most relevant works in the area published during the last 20 years. We then propose the User-Review-Item (URI) paradigm, a novel abstraction for modeling the problem that moves the focus of the feature engineering from the review to the platform level. We empirically validate the URI paradigm on a dataset of products from six Amazon categories with 270 trained models: on average, classifiers gain +4% in F1-score when considering the whole review platform context. In our experiments, we further emphasize some problems with the helpfulness prediction task: (1) the users’ writing style changes over time (i.e., concept drift), (2) past models do not generalize well across different review categories, and (3) past methods to generate the ground truth produced unreliable helpfulness scores, affecting the model evaluation phase. [ABSTRACT FROM AUTHOR]

Routing Protocol for Low-Power and Lossy Networks (RPL) is a standard routing protocol for the Low Power and Lossy Networks (LLNs). It is a part of the IPv6 over Low-Power Wireless Personal Area Network (6LoWPAN) protocol stack. Features such as energy-efficient mechanisms and availability of the secure modes of operations make RPL suitable for the constrained Internet of Things (IoT) devices. Hence, the majority of IoT applications rely on RPL for data communication. However, routing security in RPL-based IoT networks is a significant concern, motivating us to study and analyze routing attacks and suggested countermeasures against them. To this end, we provide a comprehensive survey on the state-of-the-art security threats and their corresponding countermeasures in RPL-based IoT networks. Based on our study, we propose a novel classification scheme that uses a mapping between RPL attacks and their countermeasure techniques to the RPL control messages used to develop these techniques. Furthermore, we provide an in-depth statistical analysis that includes analysis of routing attacks through the RPL control messages, distribution of various mitigation techniques as per the method used, RPL control messages involved in the mitigation techniques, and details of the tools used by multiple researchers. In the end, we highlight some open challenges and future research opportunities on this topic. We believe that this survey will be beneficial to researchers and practitioners working in the area of RPL security. [ABSTRACT FROM AUTHOR]

The challenges of cloud forensics have been well-documented by both researchers and government agencies (e.g., U.S. National Institute of Standards and Technology), although many of the challenges remain unresolved. In this article, we perform a comprehensive survey of cloud forensic literature published between January 2007 and December 2018, categorized using a five-step forensic investigation process. We also present a taxonomy of existing cloud forensic solutions, with the aim of better informing both the research and practitioner communities, as well as an in-depth discussion of existing conventional digital forensic tools and cloud-specific forensic investigation tools. Based on the findings from the survey, we present a set of design guidelines to inform future cloud forensic investigation processes, and a summary of digital artifacts that can be obtained from different stakeholders in the cloud computing architecture/ecosystem. [ABSTRACT FROM AUTHOR]

Voice-over-IP (VoIP) software are among the most widely spread and pervasive software, counting millions of monthly users. However, we argue that people ignore the drawbacks of transmitting information along with their voice, such as keystroke sounds--as such sound can reveal what someone is typing on a keyboard. In this article, we present and assess a new keyboard acoustic eavesdropping attack that involves VoIP, called Skype & Type (S&T). Unlike previous attacks, S&T assumes a weak adversary model that is very practical in many real-world settings. Indeed, S&T is very feasible, as it does not require (i) the attacker to be physically close to the victim (either in person or with a recording device) and (ii) precise profiling of the victim's typing style and keyboard; moreover, it can work with a very small amount of leaked keystrokes. We observe that leakage of keystrokes during a VoIP call is likely, as people often "multi-task" during such calls. As expected, VoIP software acquires and faithfully transmits all sounds, including emanations of pressed keystrokes, which can include passwords and other sensitive information. We show that one very popular VoIP software (Skype) conveys enough audio information to reconstruct the victim's input--keystrokes typed on the remote keyboard. Our results demonstrate that, given some knowledge on the victim's typing style and keyboard model, the attacker attains top-5 accuracy of 91.7% in guessing a random key pressed by the victim. This work extends previous results on S&T, demonstrating that our attack is effective with many different recording devices (such as laptop microphones, headset microphones, and smartphones located in proximity of the target keyboard), diverse typing styles and speed, and is particularly threatening when the victim is typing in a known language. [ABSTRACT FROM AUTHOR]

Legacy encryption systems depend on sharing a key (public or private) among the peers involved in exchanging an encrypted message. However, this approach poses privacy concerns. The users or service providers with the key have exclusive rights on the data. Especially with popular cloud services, control over the privacy of the sensitive data is lost. Even when the keys are not shared, the encrypted material is shared with a third party that does not necessarily need to access the content. Moreover, untrusted servers, providers, and cloud operators can keep identifying elements of users long after users end the relationship with the services. Indeed, Homomorphic Encryption (HE), a special kind of encryption scheme, can address these concerns as it allows any third party to operate on the encrypted data without decrypting it in advance. Although this extremely useful feature of the HE scheme has been known for over 30 years, the first plausible and achievable Fully Homomorphic Encryption (FHE) scheme, which allows any computable function to perform on the encrypted data, was introduced by Craig Gentry in 2009. Even though this was a major achievement, different implementations so far demonstrated that FHE still needs to be improved significantly to be practical on every platform. Therefore, this survey focuses on HE and FHE schemes. First, we present the basics of HE and the details of the well-known Partially Homomorphic Encryption (PHE) and Somewhat Homomorphic Encryption (SWHE), which are important pillars for achieving FHE. Then, the main FHE families, which have become the base for the other follow-up FHE schemes, are presented. Furthermore, the implementations and recent improvements in Gentry-type FHE schemes are also surveyed. Finally, further research directions are discussed. This survey is intended to give a clear knowledge and foundation to researchers and practitioners interested in knowing, applying, and extending the state-of-the-art HE, PHE, SWHE, and FHE systems. [ABSTRACT FROM AUTHOR]

The large spread of sensors and smart devices in urban infrastructures are motivating research in the area of the Internet of Things (IoT) to develop new services and improve citizens' quality of life. Sensors and smart devices generate large amounts of measurement data from sensing the environment, which is used to enable services such as control of power consumption or traffic density. To deal with such a large amount of information and provide accurate measurements, service providers can adopt information fusion, which given the decentralized nature of urban deployments can be performed by means of consensus algorithms. These algorithms allow distributed agents to (iteratively) compute linear functions on the exchanged data, and take decisions based on the outcome, without the need for the support of a central entity. However, the use of consensus algorithms raises several security concerns, especially when private or security critical information is involved in the computation. In this article we propose ODIN, a novel algorithm allowing information fusion over encrypted data. ODIN is a privacy-preserving extension of the popular consensus gossip algorithm, which prevents distributed agents from having direct access to the data while they iteratively reach consensus; agents cannot access even the final consensus value but can only retrieve partial information (e.g., a binary decision). ODIN uses efficient additive obfuscation and proxy re-encryption during the update steps and garbled circuits to make final decisions on the obfuscated consensus. We discuss the security of our proposal and show its practicability and efficiency on real-world resource-constrained devices, developing a prototype implementation for Raspberry Pi devices. [ABSTRACT FROM AUTHOR]

The popularity of Social Networking Sites (SNS) is growing rapidly, with the largest sites serving hundreds of millions of users and their private information. The privacy settings of these SNSs do not allow the user to avoid sharing some information (e.g., name and profile picture) with all the other users. Also, no matter the privacy settings, this information is always shared with the SNS (that could sell this information or be hacked). To mitigate these threats, we recently introduced the concept of Virtual Private Social Networks (VPSNs). In this work we propose the first complete architecture and implementation of VPSNs for Facebook. In particular, we address an important problem left unexplored in our previous research--that is the automatic propagation of updated profiles to all the members of the same VPSN. Furthermore, we made an in-depth study on performance and implemented several optimization to reduce the impact of VPSN on user experience. The proposed solution is lightweight, completely distributed, does not depend on the collaboration from Facebook, does not have a central point of failure, it offers (with some limitations) the same functionality as Facebook, and apart from some simple settings, the solution is almost transparent to the user. Thorough experiments, with an extended set of parameters, we have confirmed the feasibility of the proposal and have shown a very limited time-overhead experienced by the user while browsing Facebook pages. [ABSTRACT FROM AUTHOR]