Biblio

The traditional approach to distributed deep neural network (DNN) training is data-distributed learning, which partitions and distributes data to workers. This approach, although has good convergence properties, has high communication cost, which puts a strain especially on edge systems and increases delay. An emerging approach is model-distributed learning, where a training model is distributed across workers. Model-distributed learning is a promising approach to reduce communication and storage costs, which is crucial for edge systems. In this paper, we design ResPipe, a novel resilient model-distributed DNN training mechanism against delayed/failed workers. We analyze the communication cost of ResPipe and demonstrate the trade-off between resiliency and communication cost. We implement ResPipe in a real testbed consisting of Android-based smartphones, and show that it improves the convergence rate and accuracy of training for convolutional neural networks (CNNs).

In recent years, Deep Neural Networks (DNNs) have made inroads into a number of applications involving pattern recognition - from facial recognition to self-driving cars. Some of these applications, such as self-driving cars, have real-time requirements, where specialized DNN hardware accelerators help meet those requirements. Since DNN execution time is dominated by convolution, Multiply-and-Accumulate (MAC) units are at the heart of these accelerators. As hardware accelerators push the performance limits with strict power constraints, reliability is often compromised. In particular, power-constrained DNN accelerators are more vulnerable to transient and intermittent hardware faults due to particle hits, manufacturing variations, and fluctuations in power supply voltage and temperature. Methods such as hardware replication have been used to deal with these reliability problems in the past. Unfortunately, the duplication approach is untenable in a power constrained environment. This paper introduces a low-cost error-resiliency scheme that targets MAC units employed in conventional DNN accelerators. We evaluate the reliability improvements from the proposed architecture using a set of 6 CNNs over varying bit error rates (BER) and demonstrate that our proposed solution can achieve more than 99% of fault coverage with a 5-bits arithmetic code, complying with the ASIL-D level of ISO26262 standards with a negligible area and power overhead. Additionally, we evaluate the proposed detection mechanism coupled with a word masking correction scheme, demonstrating no loss of accuracy up to a BER of 10-2.

Internet of Things (IoT), Cloud and Augmented Reality (AR) are the emerging and developing technologies and are at the horizon and hype of their life cycle. Lots of commercial applications based on IoT, cloud and AR provide unrestricted access to data. The real-time applications based on these technologies are at the cusp of their innovations. The most frequent security attacks for IoT, cloud and AR applications are DDoS attacks. In this paper a detailed account of various DDoS attacks that can be the hindrance of many important sensitive services and can degrade the overall performance of recent services which are purely based on network communications. The DDoS attacks should be dealt with carefully and a set of a new generations of algorithm need to be developed to mitigate the problems caused by non-repudiation kinds of attacks.

Software and firmware vulnerabilities pose security threats to photovoltaic (PV) systems. When patches are not available or cannot be timely applied to fix vulnerabilities, it is important to mitigate vulnerabilities such that they cannot be exploited by attackers or their impacts will be limited when exploited. However, the vulnerability mitigation problem for PV systems has received little attention. This paper analyzes known security vulnerabilities in PV systems, proposes a multi-level mitigation framework and various mitigation strategies including neural network-based attack detection inside inverters, and develops a prototype system as a proof-of-concept for building vulnerability mitigation into PV system design.

With the spread of the Internet, various devices are now connected to it and the number of IoT devices is increasing. Data generated by IoT devices has traditionally been aggregated in the cloud and processed over time. However, there are two issues with using the cloud. The first is the response delay caused by the long distance between the IoT device and the cloud, and the second is the difficulty of implementing sufficient security measures on the IoT device side due to the limited resources of the IoT device at the end. To address these issues, fog computing, which is located in the middle between IoT devices and the cloud, has been attracting attention as a new network component. However, the risks associated with the introduction of fog computing have not yet been fully investigated. In this study, we conducted a risk assessment of fog computing, which is newly established to promote the use of IoT devices, and identified 24 risk factors. The main countermeasures include the gradual introduction of connected IoT connection protocols and security policy matching. We also demonstrated the effectiveness of the proposed risk measures by evaluating the risk values. The proposed risk countermeasures for fog computing should help us to utilize IoT devices in a safe and secure manner.

In recent years, new cyber attacks such as targeted attacks have caused extensive damage. With the continuing development of the IoT society, various devices are now connected to the network and are being used for various purposes. The Internet of Things has the potential to link cyber risks to actual property damage, as cyberspace risks are connected to physical space. With this increase in unknown cyber risks, the demand for cyber insurance is increasing. One of the most serious emerging risks is the silent cyber risk, and it is likely to increase in the future. However, at present, security measures against silent cyber risks are insufficient. In this study, we conducted a risk management of silent cyber risk for organizations with the objective of contributing to the development of risk management methods for new cyber risks that are expected to increase in the future. Specifically, we modeled silent cyber risk by focusing on state transitions to different risks. We newly defined two types of silent cyber risk, namely, Alteration risk and Combination risk, and conducted risk assessment. Our assessment identified 23 risk factors, and after analyzing them, we found that all of them were classified as Risk Transference. We clarified that the most effective risk countermeasure for Alteration risk was insurance and for Combination risk was measures to reduce the impact of the risk factors themselves. Our evaluation showed that the silent cyber risk could be reduced by about 50%, thus demonstrating the effectiveness of the proposed countermeasures.

We propose Cleann, the first end-to-end framework that enables online mitigation of Trojans for embedded Deep Neural Network (DNN) applications. A Trojan attack works by injecting a backdoor in the DNN while training; during inference, the Trojan can be activated by the specific backdoor trigger. What differentiates Cleann from the prior work is its lightweight methodology which recovers the ground-truth class of Trojan samples without the need for labeled data, model retraining, or prior assumptions on the trigger or the attack. We leverage dictionary learning and sparse approximation to characterize the statistical behavior of benign data and identify Trojan triggers. Cleann is devised based on algorithm/hardware co-design and is equipped with specialized hardware to enable efficient real-time execution on resource-constrained embedded platforms. Proof of concept evaluations on Cleann for the state-of-the-art Neural Trojan attacks on visual benchmarks demonstrate its competitive advantage in terms of attack resiliency and execution overhead.

3D reconstruction has piqued the interest of many disciplines, and many researchers have spent the last decade striving to improve on latest automated three-dimensional reconstruction systems. Three Dimensional models can be utilized to tackle a wide range of visualization problems as well as other activities. In this paper, we have implemented a method of Digital Surface Map (DSM) generation from Aerial images using Conditional Generative Adversarial Networks (c-GAN). We have used Seg-net architecture of Convolutional Neural Network (CNN) to segment the aerial images and then the U-net generator of c-GAN generates final DSM. The dataset we used is ISPRS Potsdam-Vaihingen dataset. We also review different stages if 3D reconstruction and how Deep learning is now being widely used to enhance the process of 3D data generation. We provide binary cross entropy loss function graph to demonstrate stability of GAN and CNN. The purpose of our approach is to solve problem of DSM generation using Deep learning techniques. We put forth our method against other latest methods of DSM generation such as Semi-global Matching (SGM) and infer the pros and cons of our approach. Finally, we suggest improvements in our methods that might be useful in increasing the accuracy.

The increased use of Convolutional Neural Networks (CNN) in mission-critical systems has increased the need for robust and resilient networks in the face of both naturally occurring faults as well as security attacks. The lack of robustness and resiliency can lead to unreliable inference results. Current methods that address CNN robustness require hardware modification, network modification, or network duplication. This paper proposes MILR a software-based CNN error detection and error correction system that enables recovery from single and multi-bit errors. The recovery capabilities are based on mathematical relationships between the inputs, outputs, and parameters(weights) of the layers; exploiting these relationships allows the recovery of erroneous parameters (iveights) throughout a layer and the network. MILR is suitable for plaintext-space error correction (PSEC) given its ability to correct whole-weight and even whole-layer errors in CNNs.

The Internet of Things (IoT) technology is being utilized in endless applications nowadays and the security of these applications is of great importance. Image based IoT applications serve a wide variety of fields such as medical application and smart cities. Steganography is a great threat to these applications where adversaries can use the images in these applications to hide malicious messages. Therefore, this paper presents an image steganalysis technique that employs Convolutional Neural Networks (CNN) to detect the infamous JPEG steganography technique: JPEG universal wavelet relative distortion (J-UNIWARD). Several experiments were conducted to determine the breaking point of J-UNIWARD, whether the hiding technique relies on correlation of the images, and the effect of utilizing Discrete Cosine Transform (DCT) on the performance of the CNN. The results of the CNN display that the breaking point of J-UNIWARD is 1.5 (bpnzAC), the correlation of the database affects the detection accuracy, and DCT increases the detection accuracy by 13%.

This paper proposes 2 multiple layer message security schemes. Information security is carried out through the implementation of cryptography, steganography and image processing techniques. In both schemes, the sensitive data is first encrypted by employing a chaotic function. In the first proposed scheme, LSB steganography is then applied to 2D slices of a 3D image. In the second proposed scheme, a corner detection filter is first applied to the 2D slices of a 3D image, then LSB embedding is carried out in those corner-detected pixels. The number of neighboring pixels used for corner detection is varied and its effect is noted. Performance of the proposed schemes is numerically evaluated using a number of metrics, including the mean squared error (MSE), the peak signal to noise ratio (PSNR), the structure similarity index measure (SSIM), the normalized cross-correlation (NCC), the image fidelity (IF), as well as the image difference (ID). The proposed schemes exhibit superior payload capacity and security in comparison to their counterparts from the literature.

Data exfiltration is of increasing concern throughout the world. The number of incidents and capabilities of data exfiltration attacks are growing at an unprecedented rate. However, such attack vectors have not been deeply explored in the literature. This paper aims to address this gap by implementing a data exfiltration methodology, detailing some data exfiltration methods. Groups of exfiltration methods are incorporated into a program that can act as a testbed for owners of any network that stores sensitive data. The implemented methods are tested against the well-known network intrusion detection system Snort, where all of them have been successfully evaded detection by its community rule sets. Thus, in this paper, we have developed new countermeasures to prevent and detect data exfiltration attempts using these methods.

A new Domain Name System (DNS) cache poisoning attack aiming at clients has emerged recently. It induced cloud users to visit fake web sites and thus reveal information such as account passwords. However, the design of current DNS defense architecture does not formally consider the protection of clients. Although the DNS traffic encryption technology can alleviate this new attack, its deployment is as slow as the new DNS architecture. Thus we propose a lightweight adaptive intelligent defense strategy, which only needs to be deployed on the client without any configuration support of DNS. Firstly, we model the attack and defense process as a static stochastic game with incomplete information under bounded rationality conditions. Secondly, to solve the problem caused by uncertain attack strategies and large quantities of game states, we adopt a deep reinforcement learning (DRL) with guaranteed monotonic improvement. Finally, through the prototype system experiment in Alibaba Cloud, the effectiveness of our method is proved against multiple attack modes with a success rate of 97.5% approximately.

Lateral movement attack performs malicious activities by infecting part of a network system first and then moving laterally to the left system in order to compromise more computers. It is widely used in various sophisticated attacks and plays a critical role. This paper aims to quantitatively analyze the transient security and dependability of a critical network system under lateral movement attacks, whose intruding capability increases with the increasing number of attacked computers. We propose a survivability model for capturing the system and adversary behaviors from the time instant of the first intrusion launched from any attacked computer to the other vulnerable computers until defense solution is developed and deployed. Stochastic Reward Nets (SRN) is applied to automatically build and solve the model. The formulas are also derived for calculating the metrics of interest. Simulation is carried out to validate the approximate accuracy of our model and formulas. The quantitative analysis can help network administrators make a trade-off between damage loss and defense cost.

Mobile Edge Computing may become a prevalent platform to support applications where mobile devices have limited compute, storage, energy and/or data privacy concerns. In this paper, we study the efficient provisioning and management of compute resources in the Edge-to-Cloud continuum for different types of real-time applications with timeliness requirements depending on application-level update rates and communication/compute delays. We begin by introducing a highly stylized network model allowing us to study the salient features of this problem including its sensitivity to compute vs. communication costs, application requirements, and traffic load variability. We then propose an online decentralized service placement algorithm, based on estimating network delays and adapting application update rates, which achieves high service availability. Our results exhibit how placement can be optimized and how a load-balancing strategy can achieve near-optimal service availability in large networks.

Network Address Translation (NAT) is an address remapping technique placed at the borders of stub domains. It is present in almost all routers and CPEs. Most NAT devices implement Port Address Translation (PAT), which allows the mapping of multiple private IP addresses to one public IP address. Based on port number information, PAT matches the incoming traffic to the corresponding "hidden" client. In an enterprise context, and with the proliferation of unauthorized wired and wireless NAT routers, NAT can be used for re-distributing an Intranet or Internet connection or for deploying hidden devices that are not visible to the enterprise IT or under its oversight, thus causing a problem known as shadow IT. Thus, it is important to detect NAT devices in an intranet to prevent this particular problem. Previous methods in identifying NAT behavior were based on features extracted from traffic traces per flow. In this paper, we propose a method to identify NAT devices using a machine learning approach from aggregated flow features. The approach uses multiple statistical features in addition to source and destination IPs and port numbers, extracted from passively collected traffic data. We also use aggregated features extracted within multiple window sizes and feed them to a machine learning classifier to study the effect of timing on NAT detection. Our approach works completely passively and achieves an accuracy of 96.9% when all features are utilized.

We propose a voting ensemble of models trained by using block-wise transformed images with secret keys against black-box attacks. Although key-based adversarial defenses were effective against gradient-based (white-box) attacks, they cannot defend against gradient-free (black-box) attacks without requiring any secret keys. In the proposed ensemble, a number of models are trained by using images transformed with different keys and block sizes, and then a voting ensemble is applied to the models. Experimental results show that the proposed defense achieves a clean accuracy of 95.56 % and an attack success rate of less than 9 % under attacks with a noise distance of 8/255 on the CIFAR-10 dataset.

The main goal of the system is to make it easier to standardize the style of program code written in C++. Based on the results of the review of existing static analyzers, in addition to the main requirements, requirements for the structure of stylistic rules were identified. Based on the results obtained, a system for static analysis of the C++ language has been developed, consisting of a set of modules. The system is implemented using the Python 3.7 programming language. HTML and CSS markup languages were used to generate html reports. To ensure that rules can be stored in the database, the MongoDB database management system and the pymongo driver module were used.

The major energy-hungry components in today's battery-operated embedded devices are mostly peripheral modules like LTE, WiFi, GPS, etc. Inefficient use of these modules causes energy hotspots, namely segments of the embedded software in which the module wastes energy. We study two such hotspots in the current paper, and provide the corresponding detection and removal algorithms based on static analysis techniques. The program code hotspots occur due to unnecessary releasing and re-acquiring of a module (which puts the module in power saving mode for a while) and misplaced acquiring of the module (which makes the module or processor to waste energy in idle mode). The detections are performed according to some relation between extreme (worst-case/best-case) execution times of some program segments and time/energy specifications of the module. The experimental results on our benchmarks show about 28 percent of energy reduction after the hotspot removals.

In this paper, physical layer security (PLS) analysis of a cooperative wireless communication system in which the source and destination nodes communicate via a relay employing decode-and-forward protocol is performed for double Rayleigh fading channel model. For the system where the source, relay and target have single antenna, an eavesdropper with multiantenna listens the source and relay together by using maximum-ratio-combining, secrecy outage and positive secrecy capacity possibilities are obtained in closed-form. The theoretical results are verified by Monte-Carlo simulations. From the results, it is observed that as the number of antennas of the eavesdropper is increased, the PLS performance of the system worsens.

Internet of Things (IoT) is becoming an emerging paradigm to provide pervasive connectivity where “anything“ can be connected “anywhere” at “anytime” via massive deployment of physical objects like sensors, controllers, and actuators. However, the open nature of wireless communications and the energy constraint of the IoT devices impose strong security concerns. In this context, traditional cryptographic techniques may not be suitable in such a resource-constrained network. To address this problem, an effective security solution that ensures a trade-off between security effectiveness and energy efficiency is required. In this paper, we exploit cooperative transmission between sensor nodes in IoT for e-Health application, as a promising technique to enhance the physical layer security of wireless communications in terms of secrecy capacity while considering the resource-impoverished devices. Specifically, we propose a dynamic and cooperative virtual multiple-input and multiple-output (MIMO) configuration approach based on game theory to preserve the confidentiality of the transmitted messages with high energy savings. For this purpose, we model the physical layer security cooperation problem as a non-transferable coalition formation game. The set of cooperative devices form a virtual dynamically-configured MIMO network that is able to securely and efficiently transmit data to the destination. Simulation results show that the proposed game-based virtual MIMO configuration approach can improve the average secrecy capacity per device as well as the network lifetime compared to non-cooperative transmission.

Phishing is an act of creating a website similar to a legitimate website with a motive of stealing user's confidential information. Phishing fraud might be the most popular cybercrime. Phishing is one of the risks that originated a couple of years back but still prevailing. This paper discusses various phishing attacks, some of the latest phishing evasion techniques used by attackers and anti-phishing approaches. This review raises awareness of those phishing strategies and helps the user to practice phishing prevention. Here, a hybrid approach of phishing detection also described having fast response time and high accuracy.

Recently, the threats of phishing attacks have in-creased. As a countermeasure against phishing attacks, phishing detection systems using deep learning techniques have been considered. However, deep learning techniques are vulnerable to adversarial examples (AEs) that intentionally cause misclassification. When AEs are applied to a deep-learning-based phishing detection system, they pose a significant security risk. Therefore, in this paper, we assess the vulnerability of a phishing detection system by inputting AEs generated based on a dataset that consists of phishing sites’ URLs. Moreover, we consider countermeasures against AEs and clarify whether these defense methods can prevent misclassification.

A successful phishing attack is prelude to various other severe attacks such as login credentials theft, unauthorized access to user’s confidential data, malware and ransomware infestation of victim’s machine etc. This paper proposes a real time lightweight machine learning based security framework for detection of phishing attacks through analysis of Uniform Resource Locators (URLs). The proposed framework initially extracts a set of highly discriminating and uncorrelated features from the URL string corpus. These extracted features are then used to transform the URL strings into their corresponding numeric feature vectors, which are eventually used to train various machine learning based classifier models for identification of malicious phishing URLs. Performance analysis of the proposed security framework on two well known datasets: Kaggle dataset and UNB dataset shows that it is capable of detecting malicious phishing URLs with high precision, while at the same time maintain a very low level of false positive rate. The proposed framework is also shown to outperform other similar security frameworks proposed in the literature.121https://www.kaggle.com/antonyj453/ur1dataset2https://www.unb.ca/cic/datasets/ur1-2016.htm1

The co-existence between Internet Protocol (IP) and Named-Data Networking (NDN) protocol is inevitable during the transition period. We propose a privacy-preserving translation method between IP and NDN called the dual-channel translation gateway. The gateway provides two different channels dedicated to the interest and the data packet to translate the IP to the NDN protocol and vice versa. Additionally, the name resolution table is provided at the gateway that binds an IP packet securely with a prefix name. Moreover, we compare the dual-channel gateway performance with the encapsulation gateway.