Biblio

Model compression is considered to be an effective way to reduce the implementation cost of deep neural networks (DNNs) while maintaining the inference accuracy. Many recent studies have developed efficient model compression algorithms and implementations in accelerators on various devices. Protecting integrity of DNN inference against fault attacks is important for diverse deep learning enabled applications. However, there has been little research investigating the fault resilience of DNNs and the impact of model compression on fault tolerance. In this work, we consider faults on different data types and develop a simulation framework for understanding the fault resiliency of compressed DNN models as compared to uncompressed models. We perform our experiments on two common DNNs, LeNet-5 and VGG16, and evaluate their fault resiliency with different types of compression. The results show that binary quantization can effectively increase the fault resilience of DNN models by 10000x for both LeNet5 and VGG16. Finally, we propose software and hardware mitigation techniques to increase the fault resiliency of DNN models.

Emergence of Information and Communication Technology (ICT) has facilitated its users to access electronic services through open channel like Internet. This approach of digital communication has its specific security lapses, which should be addressed properly to ensure Privacy, Integrity, Non-repudiation and Authentication (PINA) of information. During message communication, intruders may mount infringement attempts to compromise the communication. The situation becomes critical, if an user is identified by multiple identification numbers, as in that case, intruder have a wide window open to use any of its identification number to fulfill its ill intentions. To resolve this issue, author have proposed a single window based cloud service delivery model, where a smart card serves as a single interface to access multifaceted electronic services like banking, healthcare, employment, etc. To detect and prevent unauthorized access, in this paper, authors have focused on the intrusion detection system of the cloud service model during cloud banking transaction.

Community structure detection in social networks has become a big challenge. Various methods in the literature have been presented to solve this challenge. Recently, several methods have also been proposed to solve this challenge based on a mapping-reduction model, in which data and algorithms are divided between different process nodes so that the complexity of time and memory of community detection in large social networks is reduced. In this paper, a mapping-reduction model is first proposed to detect the structure of communities. Then the proposed framework is rewritten according to a new mechanism called distributed cache memory; distributed cache memory can store different values associated with different keys and, if necessary, put them at different computational nodes. Finally, the proposed rewritten framework has been implemented using SPARK tools and its implementation results have been reported on several major social networks. The performed experiments show the effectiveness of the proposed framework by varying the values of various parameters.

Deep neural networks are susceptible to various inference attacks as they remember information about their training data. We design white-box inference attacks to perform a comprehensive privacy analysis of deep learning models. We measure the privacy leakage through parameters of fully trained models as well as the parameter updates of models during training. We design inference algorithms for both centralized and federated learning, with respect to passive and active inference attackers, and assuming different adversary prior knowledge. We evaluate our novel white-box membership inference attacks against deep learning algorithms to trace their training data records. We show that a straightforward extension of the known black-box attacks to the white-box setting (through analyzing the outputs of activation functions) is ineffective. We therefore design new algorithms tailored to the white-box setting by exploiting the privacy vulnerabilities of the stochastic gradient descent algorithm, which is the algorithm used to train deep neural networks. We investigate the reasons why deep learning models may leak information about their training data. We then show that even well-generalized models are significantly susceptible to white-box membership inference attacks, by analyzing state-of-the-art pre-trained and publicly available models for the CIFAR dataset. We also show how adversarial participants, in the federated learning setting, can successfully run active membership inference attacks against other participants, even when the global model achieves high prediction accuracies.

Current procedures for anomaly detection in self-organizing mobile communication networks use network-centric approaches to identify dysfunctional serving nodes. In this paper, a user-centric approach and a novel methodology for anomaly detection is proposed, where the Quality of Experience (QoE) metric is used to evaluate the end-user experience. The system model demonstrates how dysfunctional serving eNodeBs are successfully detected by implementing a parametric QoE model using machine learning for prediction of user QoE in a network scenario created by the ns-3 network simulator. This approach can play a vital role in the future ultra-dense and green mobile communication networks that are expected to be both self- organizing and self-healing.

Wireless Sensor Network (WSN) faces critical security challenges due to malware(worm, virus, malicious code etc.) attack. When a single node gets compromised by malware then start to spread in entire sensor network through neighboring sensor nodes. To understand the dynamics of malware propagation in WSN proposed a Susceptible-Exposed-Infectious-Recovered-Dead (SEIRD) model. This model used the concept of epidemiology. The model focused on early detection of malicious signals presence in the network and accordingly application of security mechanism for its removal. The early detection method helps in controlling of malware spread and reduce battery consumption of sensor nodes. In this paper study the dynamics of malware propagation and stability analysis of the system. In epidemiology basic reproduction number is a crucial parameter which is used for the determination of malware status in the system. The expression of basic reproduction number has been obtained. Analyze the propagation dynamics and compared with previous model. The proposed model provides improved security mechanism in comparison to previous one. The extensive simulation results conform the analytical investigation and accuracy of proposed model.

The need for sensors to deliver, communicate, collect, alert, and share information in various applications has made wireless sensor networks very popular. However, due to its limited resources in terms of computation power, battery life and memory storage of the sensor nodes, it is challenging to add security features to provide the confidentiality, integrity, and availability. Blockchain technology ensures security and avoids the need of any trusted third party. However, applying Blockchain in a resource-constrained wireless sensor network is a challenging task because Blockchain is power, computation, and memory hungry in nature and demands heavy bandwidth due to control overheads. In this paper, a new routing and a private communication Blockchain framework is designed and tested with Constant Bit rate (CBR). The proposed Load Balancing Multi-Hop (LBMH) routing shares and enhances the battery life of the Cluster Heads and reduce control overhead during Block updates, but due to limited storage and energy of the sensor nodes, Blockchain in sensor networks may never become a reality unless computation, storage and battery life are readily available at low cost.

Computer security has gained more and more attention in a public over the last years, since computer systems are suffering from significant and increasing security threats that cause security breaches by exploiting software vulnerabilities. The most efficient way to ensure the system security is to patch the vulnerable system before a malicious attack occurs. Besides the commonly-used push-type patch management, the pull-type patch management is also adopted. The main issues in the pull-type patch management are two-fold; when to check the vulnerability information and when to apply a patch? This paper considers the security patch management for a virtual machine (VM) based intrusion tolerant system (ITS), where the system undergoes the patch management with a periodic vulnerability checking strategy, and evaluates the system security from the availability aspect. A composite stochastic reward net (SRN) model is applied to capture the attack behavior of adversary and the defense behaviors of system. Two availability measures; interval availability and point-wise availability are formulated to quantify the system security via phase expansion. The proposed approach and metrics not only enable us to quantitatively assess the system security, but also provide insights on the patch management. In numerical experiments, we evaluate effects of the intrusion rate and the number of vulnerability checking on the system security.

The presence of artificial agents in human social networks is growing. From chatbots to robots, human experience in the developed world is moving towards a socio-technical system in which agents can be technological or biological, with increasingly blurred distinctions between. Given that emotion is a key element of human interaction, enabling artificial agents with the ability to reason about affect is a key stepping stone towards a future in which technological agents and humans can work together. This paper presents work on building intelligent computational agents that integrate both emotion and cognition. These agents are grounded in the well-established social-psychological Bayesian Affect Control Theory (BayesAct). The core idea of BayesAct is that humans are motivated in their social interactions by affective alignment: they strive for their social experiences to be coherent at a deep, emotional level with their sense of identity and general world views as constructed through culturally shared symbols. This affective alignment creates cohesive bonds between group members, and is instrumental for collaborations to solidify as relational group commitments. BayesAct agents are motivated in their social interactions by a combination of affective alignment and decision theoretic reasoning, trading the two off as a function of the uncertainty or unpredictability of the situation. This paper provides a high-level view of dual process theories and advances BayesAct as a plausible, computationally tractable model based in social-psychological and sociological theory.

In this paper, development of cyber communication package in the application of grid connected solar system has been presented. Here, implemented communication methodology supports communication process with reduced latency, high security arrangement with various degrees of freedom. Faithful transferring of various electrical data for the purpose of measurement, monitoring and controlling actions depend on the bidirectional communication strategy. Thus, real-time communication of data through cyber network has been emphasized in this paper. The C\# language based coding is done to develop the communication program. The notable features of proposed communication process are reduction of latency during data exchange by usage of advanced encryption standard (AES) algorithm, tightening of cyber security arrangement by implementing secured socket layer (SSL) and Rivest, Shamir and Adleman (RSA) algorithms. Various real-time experiments using internet connected computers have been done to verify the usability of the proposed communication concept along with its notable features in the application.

Machine learning has been adopted widely to perform prediction and classification. Implementing machine learning increases security risks when computation process involves sensitive data on training and testing computations. We present a proposed system to protect machine learning engines in IoT environment without modifying internal machine learning architecture. Our proposed system is designed for passwordless and eliminated the third-party in executing machine learning transactions. To evaluate our a proposed system, we conduct experimental with machine learning transactions on IoT board and measure computation time each transaction. The experimental results show that our proposed system can address security issues on machine learning computation with low time consumption.

Network security became a viral topic nowadays, Anomaly-based Intrusion Detection Systems [1] (IDSs) plays an indispensable role in identifying the attacks from networks and the detection rate and accuracy are said to be high. The proposed work explore this topic and solve this issue by the IDS model developed using Artificial Neural Network (ANN). This model uses Feed - Forward Neural Net algorithms and Probabilistic Neural Network and oppositional based on Particle Swarm optimization Algorithm for lessen the computational overhead and boost the performance level. The whole computing overhead produced in its execution and training are get minimized by the various optimization techniques used in these developed ANN-based IDS system. The experimental study on the developed system tested using the standard NSL-KDD dataset performs well, while compare with other intrusion detection models, built using NN, RB and OPSO algorithms.

Recent years have seen an increased interest in digital wallets for a multitude of use cases including online banking, cryptocurrency, and digital identity management. Digital wallets play a pivotal role in the secure management of cryptographic keys and credentials, and for providing certain identity management services. In this paper, we examine a proof-of-concept digital wallet in the context of Self-Sovereign Identity and provide a practical decentralized key recovery solution using Shamir's secret sharing scheme and Hyperledger Indy distributed ledger technology.

By accurately measuring risk for enterprise networks, attack graphs allow network defenders to understand the most critical threats and select the most effective countermeasures. This paper describes substantial enhancements to the NetSPA attack graph system required to model additional present-day threats (zero-day exploits and client-side attacks) and countermeasures (intrusion prevention systems, proxy firewalls, personal firewalls, and host-based vulnerability scans). Point-to-point reachability algorithms and structures were extensively redesigned to support "reverse" reachability computations and personal firewalls. Host-based vulnerability scans are imported and analyzed. Analysis of an operational network with 84 hosts demonstrates that client-side attacks pose a serious threat. Experiments on larger simulated networks demonstrated that NetSPA's previous excellent scaling is maintained. Less than two minutes are required to completely analyze a four-enclave simulated network with more than 40,000 hosts protected by personal firewalls.

In this study, a systematic mapping study was conducted to systematically evaluate publications on Intrusion Detection Systems with Deep Learning. 6088 papers have been examined by using systematic mapping method to evaluate the publications related to this paper, which have been used increasingly in the Intrusion Detection Systems. The goal of our study is to determine which deep learning algorithms were used mostly in the algortihms, which criteria were taken into account for selecting the preferred deep learning algorithm, and the most searched topics of intrusion detection with deep learning algorithm model. Scientific studies published in the last 10 years have been studied in the IEEE Explorer, ACM Digital Library, Science Direct, Scopus and Wiley databases.

An adaptable agent-based IDS (AAIDS) inspired by the danger theory of artificial immune system is proposed. The learning mechanism of AAIDS is designed by emulating how dendritic cells (DC) in immune systems detect and classify danger signals. AG agent, DC agent and TC agent coordinate together and respond to system calls directly rather than analyze network packets. Simulations show AAIDS can determine several critical scenarios of the system behaviors where packet analysis is impractical.

Many enterprises and individuals prefer to outsource data to public cloud via various pricing approaches. One of the most widely-used approaches is the pay-as-you-go model, where the data owner hires public cloud to share data with data consumers, and only pays for the actually consumed services. To realize controllable and secure data sharing, ciphertext-policy attribute-based encryption (CP-ABE) is a suitable solution, which can provide fine-grained access control and encryption functionalities simultaneously. But there are some serious challenges when applying CP-ABE in pay-as-you-go. Firstly, the decryption cost in ABE is too heavy for data consumers. Secondly, ABE ciphertexts probably suffer distributed denial of services (DDoS) attacks, but there is no solution that can eliminate the security risk. At last, the data owner should audit resource consumption to guarantee the transparency of charge, while the existing method is inefficient. In this work, we propose a general construction named fully accountable ABE (FA-ABE), which simultaneously solves all the challenges by supporting all-sided accountability in the pay-as-you-go model. We formally define the security model and prove the security in the standard model. Also, we implement an instantiate construction with the self-developed library libabe. The experiment results indicate the efficiency and practicality of our construction.

DNA cryptography becomes a burgeoning new area of study along with the fast-developing of DNA computing and modern cryptography. Point-doubling, point-addition and point-multiplication are three fundamental point-operations to construct encryption protocols in some cryptosystem over mathematical curves such as elliptic curves and conic curves. This paper proposes a DNA computing model to calculate point-doubling in conic curves cryptosystem over finite held GF(2n). By decomposing and rearranging the computing steps of point-doubling, the assembly process could be fulfilled by using 8 different types of computation tiles performing different functions with 1097 encoding ways. This model could also figure out point-multiplication if its coefficient is 2k. The assembly time complexity is 2kn+n-k-1, and the space complexity is k2n2+kn2-k2n.

In this paper, We propose DNA sticker model based algorithm, a computability model, which is a simulation of the parallel computations using the Molecular computing as in Adelman's DNA computing experiment, it demonstrates how to use a sticker-based model to design a simple DNA-based algorithm for attacking a linear and a non-linear feedback shift register (FSR) based stream cipher. The algorithm first construct the TEST TUBE contains all overall solution space of memory complexes for the cipher and initials of registers via the sticker-based model. Then, with biological operations, separate and combine, we remove those which encode illegal plain and key stream from the TEST TUBE of memory complexes, the decision based on verifying a key stream bit this bit represented by output of LFSRs equation. The model anticipates two basic groups of single stranded DNA molecules in its representation one of a genetic bases and second of a bit string, It invests parallel search into the space of solutions through the possibilities of DNA computing and makes use of the method of cryptanalysis of algebraic code as a decision technique to accept the solution or not, and their operations are repeated until one solution or limited group of solutions is reached. The main advantages of the suggested algorithm are limited number of cipher characters, and finding one exact solution The present work concentrates on showing the applicability of DNA computing concepts as a powerful tool in breaking cryptographic systems.

Fog-Cloud framework is being regarded as a more promising technology to provide performance guarantee for IoT applications, which not only have higher requirements on computation resources, but also are delay and/or security sensitive. In this framework, a delay and security-sensitive computation task is usually divided into several sub-tasks, which could be offloaded to either fog or cloud computing servers, referred to as offloading destinations. Sub-tasks may exchange information during their processing and then have requirement on transmission bandwidth. Different destinations produce different completion delays of a sub-task, affecting the corresponding task delay. The existing offloading approaches either considered only a single type of offloading destinations or ignored delay and/or security constraint. This paper studies a computation offloading problem in the fog-cloud scenario where not only computation and security capabilities of offloading destinations may be different, but also bandwidth and delay of links may be different. We first propose a joint offloading approach by formulating the problem as a form of Mixed Integer Programming Multi-Commodity Flow to maximize the fog-cloud provider's revenue without sacrificing performance and security requirements of users. We also propose a greedy algorithm for the problem. Extensive simulation results under various network scales show that the proposed computation offloading mechanism achieves higher revenue than the conventional single-type computation offloading under delay and security constraints.

Fog and Edge computing provide a large pool of resources at the edge of the network that may be used for distributed computing. Fog infrastructure heterogeneity also results in complex configuration of distributed applications on computing nodes. Linux containers are a mainstream technique allowing to run packaged applications and micro services. However, running applications on remote hosts owned by third parties is challenging because of untrusted operating systems and hardware maintained by third parties. To meet such challenges, we may leverage trusted execution mechanisms. In this work, we propose a model for distributed computing on Fog infrastructures using Linux containers secured by Intel's Software Guard Extensions (SGX) technology. We implement our model on a Docker and OpenSGX platform. The result is a secure and flexible approach for distributed computing on Fog infrastructures.

This work proposes an approach to find and learn informative representations from 2 dimensional gray-level images for facial expression recognition application. The learned features are obtained from a designed convolutional neural network (CNN). The developed CNN enables us to learn features from the images in a highly efficient manner by cascading different layers together. The developed model is computationally efficient since it does not consist of a huge number of layers and at the same time it takes into consideration the overfitting problem. The outcomes from the developed CNN are compared to handcrafted features that span texture and shape features. The experiments conducted on the Bosphours database show that the developed CNN model outperforms the handcrafted features when coupled with a Support Vector Machines (SVM) classifier.

Facial expression is the most powerful and natural non-verbal emotional communication method. Facial Expression Recognition(FER) has significance in machine learning tasks. Deep Learning models perform well in FER tasks, but it doesn't provide any justification for its decisions. Based on the hypothesis that facial expression is a combination of facial muscle movements, we find that Facial Action Coding Units(AUs) and Emotion label have a relationship in CK+ Dataset. In this paper, we propose a model which utilises AUs to explain Convolutional Neural Network(CNN) model's classification results. The CNN model is trained with CK+ Dataset and classifies emotion based on extracted features. Explanation model classifies the multiple AUs with the extracted features and emotion classes from the CNN model. Our experiment shows that with only features and emotion classes obtained from the CNN model, Explanation model generates AUs very well.

The network attack graph is a powerful tool for analyzing network security, but the generation of a large-scale graph is non-trivial. The main challenge is from the explosion of network state space, which greatly increases time and storage costs. In this paper, three parallel algorithms are proposed to generate scalable attack graphs. An OpenMP-based programming implementation is used to test their performance. Compared with the serial algorithm, the best performance from the proposed algorithms provides a 10X speedup.

We investigate a method of building a closed domain intelligent conversational agent using deep neural networks. A conversational agent is a dialog system intended to converse with a human, with a coherent structure. Our conversational agent uses a retrieval based model that identifies the intent of the input user query and maps it to a knowledge base to return appropriate results. Human conversations are based on context, but existing conversational agents are context insensitive. To overcome this limitation, our system uses a simple stack based context identification and storage system. The conversational agent generates responses according to the current context of conversation. allowing more human-like conversations.