Biblio

Attribution methods can provide powerful insights into the reasons for a classifier's decision. We argue that a key desideratum of an explanation method is its robustness to input hyperparameters which are often randomly set or empirically tuned. High sensitivity to arbitrary hyperparameter choices does not only impede reproducibility but also questions the correctness of an explanation and impairs the trust of end-users. In this paper, we provide a thorough empirical study on the sensitivity of existing attribution methods. We found an alarming trend that many methods are highly sensitive to changes in their common hyperparameters e.g. even changing a random seed can yield a different explanation! Interestingly, such sensitivity is not reflected in the average explanation accuracy scores over the dataset as commonly reported in the literature. In addition, explanations generated for robust classifiers (i.e. which are trained to be invariant to pixel-wise perturbations) are surprisingly more robust than those generated for regular classifiers.

Self-sovereign identities provide user autonomy and immutability to individual identities and full control to their identity owners. The immutability and control are possible by implementing identities in a decentralized manner on blockchains that are specially designed for identity operations such as Hyperledger Indy. As with any type of identity, self-sovereign identities too deal with Personally Identifiable Information (PII) of the identity holders and comes with the usual risks of privacy and security. This study examined certain scenarios of personal data disclosure via credential exchanges between such identities and risks of man-in-the-middle attacks in the blockchain based identity system Hyperledger Indy. On the basis of the findings, the paper proposes the following enhancements: 1) A novel attribute sensitivity score model for self-sovereign identity agents to ascertain the sensitivity of attributes shared in credential exchanges 2) A method of mitigating man-in-the-middle attacks between peer self-sovereign identities and 3) A novel quantitative model for determining a credential issuer's reputation based on the number of issued credentials in a window period, which is then utilized to calculate an overall confidence level score for the issuer.

With the rapid progress of informatization construction in power business, data resource has become the basic strategic resource of the power industry and innovative element in power production. The security protection of data in power business is particularly important in the informatization construction of power business. In order to implement data security protection, transparent encryption is one of the fifteen key technical standards in the Construction Guideline of the Standard Network Data Security System. However, data storage in the encrypted state is bound to affect the security audit of data to a certain extent. Based on this problem, this paper proposes a scheme to audit the sensitivity of the power business data under the protection of encryption to achieve an efficient sensitivity audit of ciphertext data with the premise of not revealing the decryption key or data information. Through a security demonstration, this paper fully proves that this solution is secure under the known plaintext attacks.

Botnets are one of the major threats on the Internet. They are used for malicious activities to compromise the basic network security goals, namely Confidentiality, Integrity, and Availability. For reliable botnet detection and defense, deep learning-based approaches were recently proposed. In this paper, four different deep learning models, namely Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), hybrid CNN-LSTM, and Multi-layer Perception (MLP) are applied for botnet detection and simulation studies are carried out using the CTU-13 botnet traffic dataset. We use several performance metrics such as accuracy, sensitivity, specificity, precision, and F1 score to evaluate the performance of each model on classifying both known and unknown (zero-day) botnet traffic patterns. The results show that our deep learning models can accurately and reliably detect both known and unknown botnet traffic, and show better performance than other deep learning models.

One of the main threats to the underwater communication cables identified in the recent years is possible tampering or damage by malicious actors. This paper proposes a solution with explanation abilities to detect and investigate this kind of threat within the evidence theory framework. The reasoning scheme implements the traditional “opportunity-capability-intent” threat model to assess a degree to which a given vessel may pose a threat. The scenario discussed considers a variety of possible pieces of information available from different sources. A source quality model is used to reason with the partially reliable sources and the impact of this meta-information on the overall assessment is illustrated. Examples of uncertain relationships between the relevant variables are modelled and the constructed model is used to investigate the probability of threat of four vessels of different types. One of these cases is discussed in more detail to demonstrate the explanation abilities. Explanations about inference are provided thanks to sensitivity spaces in which the impact of the different pieces of information on the reasoning are compared.

Due to the noise in the images, the edges extracted from these noisy images are always discontinuous and inaccurate by traditional operators. In order to solve these problems, this paper proposes multi-direction edge detection operator to detect edges from noisy images. The new operator is designed by introducing the shear transformation into the traditional operator. On the one hand, the shear transformation can provide a more favorable treatment for directions, which can make the new operator detect edges in different directions and overcome the directional limitation in the traditional operator. On the other hand, all the single pixel edge images in different directions can be fused. In this case, the edge information can complement each other. The experimental results indicate that the new operator is superior to the traditional ones in terms of the effectiveness of edge detection and the ability of noise rejection.

Fully homomorphic encryption (FHE) was one of the most prominent research topics of the last ten years. And it is considered as a major cryptographic tool in a secure and reliable cloud computing environment. The reason behind that because it allows computations over encrypted data, without decrypting the original message. This paper developed a new symmetric (FHE) algorithm based on Enhanced Matrix Operation for Randomization and Encryption (EMORE) algorithm using a chaotic system. The proposed algorithm was considered a noise-free algorithm. It generates the ciphertext in a floating-point number's format, overcomes the problem of plaintext ring and modular arithmetic operation in EMORE by the hardness of a chaotic system, and provides another level of security in terms of randomness properties, sensitivity to the initial condition, and large key size (\textbackslashtextgreater2100) of a chaotic system. Besides that, the proposed algorithm provides the confidentiality and privacy of outsourced data computing through homomorphism property of it. By using both numerical and statistical tests, these tests proved that the proposed algorithm has positive randomness properties and provide secure and reliable encryption (through encryption-decryption time, key sensitivity, keyspace, and correlation). Finally, the simulation results show that the execution time of the proposed algorithm is faster about 7.85 times than the EMORE algorithm.

Sybil attacks are one of the most prominent security problems of trust mechanisms in a distributed network with a large number of highly dynamic and heterogeneous devices, which expose serious threat to edge computing based distributed systems. Graphbased Sybil detection approaches extract social structures from target distributed systems, refine the graph via preprocessing methods and capture Sybil nodes based on the specific properties of the refined graph structure. Graph preprocessing is a critical component in such Sybil detection methods, and intuitively, the processing methods will affect the detection performance. Thoroughly understanding the dependency on the graph-processing methods is very important to develop and deploy Sybil detection approaches. In this paper, we design experiments and conduct systematic analysis on graph-based Sybil detection with respect to different graph preprocessing methods on selected network environments. The experiment results disclose the sensitivity caused by different graph transformations on accuracy and robustness of Sybil detection methods.

This paper revealed the development and implementation of the wearable sensors based on transient responses of textile chemical sensors for odorant detection system as wearable sensor of humanoid robot. The textile chemical sensors consist of nine polymer/CNTs nano-composite gas sensors which can be divided into three different prototypes of the wearable humanoid robot; (i) human axillary odor monitoring, (ii) human foot odor tracking, and (iii) wearable personal gas leakage detection. These prototypes can be integrated into high-performance wearable wellness platform such as smart clothes, smart shoes and wearable pocket toxic-gas detector. While operating mode has been designed to use ZigBee wireless communication technology for data acquisition and monitoring system. Wearable humanoid robot offers several platforms that can be applied to investigate the role of individual scent produced by different parts of the human body such as axillary odor and foot odor, which have potential health effects from abnormal or offensive body odor. Moreover, wearable personal safety and security component in robot is also effective for detecting NH3 leakage in environment. Preliminary results with nine textile chemical sensors for odor biomarker and NH3 detection demonstrates the feasibility of using the wearable humanoid robot to distinguish unpleasant odor released when you're physically active. It also showed an excellent performance to detect a hazardous gas like ammonia (NH3) with sensitivity as low as 5 ppm.

Malware analysis and detection based on it is very important factor in the computer security. Despite of the enormous effort of companies making anti-malware solutions, it is usually not possible to respond to new malware in time and some computers will get infected. This shortcoming could be partially mitigated through using behavioral malware analysis. This work is aimed towards machine learning algorithms comparison for the behavioral malware analysis purposes.

Information communicated within cyber-physical systems (CPSs) is often used in determining the physical states of such systems, and malicious adversaries may intercept these communications in order to infer future states of a CPS or its components. Accordingly, there arises a need to protect the state values of a system. Recently, the notion of differential privacy has been used to protect state trajectories in dynamical systems, and it is this notion of privacy that we use here to protect the state trajectories of CPSs. We incorporate a cloud computer to coordinate the agents comprising the CPSs of interest, and the cloud offers the ability to remotely coordinate many agents, rapidly perform computations, and broadcast the results, making it a natural fit for systems with many interacting agents or components. Striving for broad applicability, we solve infinite-horizon linear-quadratic-regulator (LQR) problems, and each agent protects its own state trajectory by adding noise to its states before they are sent to the cloud. The cloud then uses these state values to generate optimal inputs for the agents. As a result, private data are fed into feedback loops at each iteration, and each noisy term affects every future state of every agent. In this paper, we show that the differentially private LQR problem can be related to the well-studied linear-quadratic-Gaussian (LQG) problem, and we provide bounds on how agents' privacy requirements affect the cloud's ability to generate optimal feedback control values for the agents. These results are illustrated in numerical simulations.

The increasingly networked human society requires that human beings have a clear understanding and control over the structure, nature and behavior of various social networks. There is a tendency towards privacy in the study of network evolutions because privacy disclosure behavior in the network has gradually developed into a serious concern. For this purpose, we extended information theory and proposed a brand-new concept about so-called “habitual privacy” to quantitatively analyze privacy exposure behavior and facilitate privacy computation. We emphasized that habitual privacy is an inherent property of the user and is correlated with their habitual behaviors. The widely approved driving force in recent modeling complex networks is originated from activity. Thus, we propose the privacy-driven model through synthetically considering the activity impact and habitual privacy underlying the decision process. Privacy-driven model facilitates to more accurately capture highly dynamical network behaviors and figure out the complex evolution process, allowing a profound understanding of the evolution of network driven by privacy.

The performance of the trust evaluation strategy depends on the accuracy and rationality of the trust evaluation weight system. Trust is a difficult to accurate measurement and quantitative cognition in the heart, the trust of the traditional evaluation method has a strong subjectivity and fuzziness and uncertainty. This paper uses the AHP method to determine the trust evaluation index weight, and combined with grey system theory to build trust gray evaluation model. The use of gray assessment based on the whitening weight function in the evaluation process reduces the impact of the problem that the evaluation result of the trust evaluation is not easy to accurately quantify when the decision fuzzy and the operating mechanism are uncertain.

With the exponential growth in the usage of electronic medical records (EMR), the amount of data generated by the healthcare industry has too increased exponentially. These large amounts of data, known as “Big Data” is mostly unstructured. Special big data analytics methods are required to process the information and retrieve information which is meaningful. As patient information in hospitals and other healthcare facilities become increasingly electronic, Big Data technologies are needed now more than ever to manage and understand this data. In addition, this information tends to be quite sensitive and needs a highly secure environment. However, current security algorithms are hard to be implemented because it would take a huge amount of time and resources. Security protocols in Big data are also not adequate in protecting sensitive information in the healthcare. As a result, the healthcare data is both heterogeneous and insecure. As a solution we propose the Secure Pattern-Based Data Sensitivity Framework (PBDSF), that uses machine learning mechanisms to identify the common set of attributes of patient data, data frequency, various patterns of codes used to identify specific conditions to secure sensitive information. The framework uses Hadoop and is built on Hadoop Distributed File System (HDFS) as a basis for our clusters of machines to process Big Data, and perform tasks such as identifying sensitive information in a huge amount of data and encrypting data that are identified to be sensitive.

Distributed Denial of Services (DDoS) attacks continue to be one of the most challenging threats to the Internet. The intensity and frequency of these attacks are increasing at an alarming rate. Numerous schemes have been proposed to mitigate the impact of DDoS attacks. This paper presents a comprehensive empirical evaluation of Machine Learning (ML)based DDoS detection techniques, to gain better understanding of their performance in different types of environments. To this end, a framework is developed, focusing on different attack scenarios, to investigate the performance of a class of ML-based techniques. The evaluation uses different performance metrics, including the impact of the “Class Imbalance Problem” on ML-based DDoS detection. The results of the comparative analysis show that no one technique outperforms all others in all test cases. Furthermore, the results underscore the need for a method oriented feature selection model to enhance the capabilities of ML-based detection techniques. Finally, the results show that the class imbalance problem significantly impacts performance, underscoring the need to address this problem in order to enhance ML-based DDoS detection capabilities.

Differential privacy is heavily used in privacy protection due to it provides strong protection against private data. The existing differential privacy scheme mainly implements the privacy protection of nodes or edges in the network by perturbing the data query results. Most of them cannot meet the privacy protection requirements of multiple types of information. In order to overcome these issues, a differential privacy security mechanism with average path length (APL) query is proposed in this paper, which realize the privacy protection of both network vertices and edge weights. Firstly, by describing APL, the reasons for choosing this attribute as the query function are analyzed. Secondly, global sensitivity of APL query under the need of node privacy protection and edge-weighted privacy protection is proved. Finally, the relationship between data availability and privacy control parameters in differential privacy is analyzed through experiments.

Deep learning is a popular powerful machine learning solution to the computer vision tasks. The most criticized vulnerability of deep learning is its poor tolerance towards adversarial images obtained by deliberately adding imperceptibly small perturbations to the clean inputs. Such negatives can delude a classifier into wrong decision making. Previous defensive techniques mostly focused on refining the models or input transformation. They are either implemented only with small datasets or shown to have limited success. Furthermore, they are rarely scrutinized from the hardware perspective despite Artificial Intelligence (AI) on a chip is a roadmap for embedded intelligence everywhere. In this paper we propose a new discriminative noise injection strategy to adaptively select a few dominant layers and progressively discriminate adversarial from benign inputs. This is made possible by evaluating the differences in label change rate from both adversarial and natural images by injecting different amount of noise into the weights of individual layers in the model. The approach is evaluated on the ImageNet Dataset with 8-bit truncated models for the state-of-the-art DNN architectures. The results show a high detection rate of up to 88.00% with only approximately 5% of false positive rate for MobileNet. Both detection rate and false positive rate have been improved well above existing advanced defenses against the most practical noninvasive universal perturbation attack on deep learning based AI chip.

Privacy preservation on social networks is nowadays a societal issue. In this paper, our contributions are to establish such a model for privacy preservation. We use differential privacy for personal privacy analysis and measurement. Our conclusion is that privacy could be measured and preserved if the corresponding approaches could be taken.

Privacy preservation on social networks is nowadays a societal issue. In this paper, our contributions are to establish such a model for privacy preservation. We use differential privacy for personal privacy analysis and measurement. Our conclusion is that privacy could be measured and preserved if the corresponding approaches could be taken.

Security and consistency of smart grids is one of the main issues in the design and maintenance of highly controlled and monitored new power grids. Bad data injection attack could lead to disasters such as power system outage, or huge economical losses. In many attack scenarios, the attacker can come up with new attack strategies that couldn't be detected by the traditional bad data detection methods. Adaptive Partitioning State Estimation (APSE) method [3] has been proposed recently to combat such attacks. In this work, we evaluate and compare with a traditional method. The main idea of APSE is to increase the sensitivity of the chi-square test by partitioning the large grids into small ones and apply the test on each partition individually and repeat this procedure until the faulty node is located. Our simulation findings using MATPOWER program show that the method is not consistent where it is sensitive the systems size and the location of faulty nodes as well.

keystroke dynamics authenticates the system user by analyzing his typing rhythm. Given that each of us has his own typing rhythm and that the method is based on the keyboard makes it available in all computer machines, these two reasons (uniqueness and reduced cost) have made the method very solicit by administrators of security. In addition, the researchers used the method in different fields that are listed later in the paper.

Surface acoustic wave (SAW) optical sensor based on two-dimensional (2D) sensing layer can always provide extremely high sensitivity. As an attractive option, the application of exfoliated 2D perovskite on acousto-optic coupling optical sensor is investigated. In this work, exfoliated 2D (PEA)2PbI4 sheet was transferred as a sensing layer onto the delay area of a dual-port SAW resonator with resonant frequency 497 MHz. From the response under 532 nm laser with intensity of 0.9 mW/cm2, a largest frequency shift of 13.92 MHz was observed. The ultrahigh sensitivity up to 31.6 ppm/(μW/cm2) was calculated by experiment results. We also carried out theoretical analysis and finite element simulation of 3D model to demonstrate the mechanism and validity for optical sensing. The fabricated optical sensor expressed great potential for a variety of optical applications.

How to protect users' private data during network data mining has become a hot issue in the fields of big data and network information security. Most current researches on differential privacy k-means clustering algorithms focus on optimizing the selection of initial centroids. However, the traditional privacy budget allocation has the problem that the random noise becomes too large as the number of iterations increases, which will reduce the performance of data clustering. To solve the problem, we improved the way of privacy budget allocation in differentially private clustering algorithm DPk-means, and proposed APDPk-means, a new differential privacy clustering algorithm based on arithmetic progression privacy budget allocation. APDPk-means decomposes the total privacy budget into a decreasing arithmetic progression, allocating the privacy budgets from large to small in the iterative process, so as to ensure the rapid convergence in early iteration. The experiment results show that compared with the other differentially private k-means algorithms, APDPk-means has better performance in availability and quality of the clustering result under the same level of privacy protection.

Data have become an important asset for analysis and behavioral prediction, especially correlations between data. Privacy protection has aroused academic and social concern given the amount of personal sensitive information involved in data. However, existing works assume that the records are independent of each other, which is unsuitable for associated data. Many studies either fail to achieve privacy protection or lead to excessive loss of information while applying data correlations. Differential privacy, which achieves privacy protection by injecting random noise into the statistical results given the correlation, will improve the background knowledge of adversaries. Therefore, this paper proposes an information entropy differential privacy solution for correlation data privacy issues based on rough set theory. Under the solution, we use rough set theory to measure the degree of association between attributes and use information entropy to quantify the sensitivity of the attribute. The information entropy difference privacy is achieved by clustering based on the correlation and adding personalized noise to each cluster while preserving the correlations between data. Experiments show that our algorithm can effectively preserve the correlation between the attributes while protecting privacy.

The sensitivity of a function is the maximum change of its output for a unit change of its input. In this paper we present a method for determining the sensitivity of SQL queries, seen as functions from databases to datasets, where the change is measured in the number of rows that differ. Given a query, a database schema and a number, our method constructs a formula that is satisfiable only if the sensitivity of the query is bigger than this number. Our method is composable, and can also be applied to SQL workflows. Our results can be used to calibrate the amount of noise that has to be added to the output of the query to obtain a certain level of differential privacy.