Biblio

With the ever-growing occurrence of networking attacks, robust network security systems are essential to prevent and mitigate their harming effects. In recent years, machine learning-based systems have gain popularity for network security applications, usually considering the application of shallow models, where a set of expert handcrafted features are needed to pre-process the data before training. The main problem with this approach is that handcrafted features can fail to perform well given different kinds of scenarios and problems. Deep Learning models can solve this kind of issues using their ability to learn feature representations from input raw or basic, non-processed data. In this paper we explore the power of deep learning models on the specific problem of detection and classification of malware network traffic, using different representations for the input data. As a major advantage as compared to the state of the art, we consider raw measurements coming directly from the stream of monitored bytes as the input to the proposed models, and evaluate different raw-traffic feature representations, including packet and flow-level ones. Our results suggest that deep learning models can better capture the underlying statistics of malicious traffic as compared to classical, shallow-like models, even while operating in the dark, i.e., without any sort of expert handcrafted inputs.

False data injection is an on-going concern facing power system state estimation. In this work, a neural network is trained to detect the existence of false data in measurements. The proposed approach can make use of historical data, if available, by using them in the training sets of the proposed neural network model. However, the inputs of perceptron model in this work are the residual elements from the state estimation, which are highly correlated. Therefore, their dimension could be reduced by preserving the most informative features from the inputs. To this end, principal component analysis is used (i.e., a data preprocessing technique). This technique is especially efficient for highly correlated data sets, which is the case in power system measurements. The results of different perceptron models that are proposed for detection, are compared to a simple perceptron that produces identical result to the outlier detection scheme. For generating the training sets, state estimation was run for different false data on different measurements in 13-bus IEEE test system, and the residuals are saved as inputs of training sets. The testing results of the trained network show its good performance in detection of false data in measurements.

We investigate how to protect features corresponding to personal information using homomorphic encryption when matching people in several camera views. Homomorphic encryption can compute a distance between features without decryption. Thus, our method is able to use a computing server on a public network while protecting personal information. To apply homomorphic encryption, our method uses linear quantization to represent each element of the feature as integers. Experimental results show that there is no significant difference in the accuracy of person re-identification with or without homomorphic encryption and linear quantization.

In this paper, we propose a method for improving the robustness of real-time facial expression recognition. Although there are many ways to improve the accuracy of facial expression recognition, a revamp of the training framework and image preprocessing allow better results in applications. One existing problem is that when the camera is capturing images in high speed, changes in image characteristics may occur at certain moments due to the influence of light and other factors. Such changes can result in incorrect recognition of the human facial expression. To solve this problem for smooth system operation and maintenance of recognition speed, we take changes in image characteristics at high speed capturing into account. The proposed method does not use the immediate output for reference, but refers to the previous image for averaging to facilitate recognition. In this way, we are able to reduce interference by the characteristics of the images. The experimental results show that after adopting this method, overall robustness and accuracy of facial expression recognition have been greatly improved compared to those obtained by only the convolution neural network (CNN).

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 emotion recognition is an essential aspect in human-machine interaction. In the real-world conditions, it faces many challenges, i.e., illumination changes, large pose variations and partial or full occlusions, which cause different facial areas with different sharpness and completeness. Inspired by this fact, we focus on facial expression recognition based on partial faces in this paper. We compare contribution of seven facial areas of low-resolution images, including nose areas, mouse areas, eyes areas, nose to mouse areas, nose to eyes areas, mouth to eyes areas and the whole face areas. Through analysis on the confusion matrix and the class activation map, we find that mouth regions contain much emotional information compared with nose areas and eyes areas. In the meantime, considering larger facial areas is helpful to judge the expression more precisely. To sum up, contributions of this paper are two-fold: (1) We reveal concerned areas of human in emotion recognition. (2) We quantify the contribution of different facial parts.

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.

Everyday., the DoS/DDoS attacks are increasing all over the world and the ways attackers are using changing continuously. This increase and variety on the attacks are affecting the governments, institutions, organizations and corporations in a bad way. Every successful attack is causing them to lose money and lose reputation in return. This paper presents an introduction to a method which can show what the attack and where the attack based on. This is tried to be achieved with using clustering algorithm DBSCAN on network traffic because of the change and variety in attack vectors.

Software Defined Network (SDN) is a revolutionary idea to realize software-driven network with the separation of control and data planes. In essence, SDN addresses the problems faced by the traditional network architecture; however, it may as well expose the network to new attacks. Among other attacks, distributed denial of service (DDoS) attacks are hard to contain in such software-based networks. Existing DDoS mitigation techniques either lack in performance or jeopardize the accuracy of the attack detection. To fill the voids, we propose in this paper a machine learning-based DDoS mitigation technique for SDN. First, we create a model for DDoS detection in SDN using NSL-KDD dataset and then after training the model on this dataset, we use real DDoS attacks to assess our proposed model. Obtained results show that the proposed technique equates favorably to the current techniques with increased performance and accuracy.

Hiring seasonal workers in call centers to provide customer service is a common practice in B2C companies. The quality of service delivered by both contracting and employee customer service agents depends heavily on the domain knowledge available to them. When observing the internal group messaging channels used by agents, we found that similar questions are often asked repetitively by different agents, especially from less experienced ones. The goal of our work is to leverage the promising advances in conversational AI to provide a chatbot-like mechanism for assisting agents in promptly resolving a customer's issue. In this paper, we develop a neural-based conversational solution that employs BiLSTM with attention mechanism and demonstrate how our system boosts the effectiveness of customer support agents. In addition, we discuss the design principles and the necessary considerations for our system. We then demonstrate how our system, named "Isa" (Intuit Smart Agent), can help customer service agents provide a high-quality customer experience by reducing customer wait time and by applying the knowledge accumulated from customer interactions in future applications.

An important ingredient for a successful recipe for solving machine learning problems is the availability of a suitable dataset. However, such a dataset may have to be extracted from a large unstructured and semi-structured data like programming code, scripts, and text. In this work, we propose a plug-in based, extensible feature extraction framework for which we have prototyped as a tool. The proposed framework is demonstrated by extracting features from two different sources of semi-structured and unstructured data. The semi-structured data comprised of web page and script based data whereas the other data was taken from email data for spam filtering. The usefulness of the tool was also assessed on the aspect of ease of programming.

This paper studies the principle of vulnerability generation and mechanism of cross-site scripting attack, designs a dynamic cross-site scripting vulnerabilities detection technique based on existing theories of black box vulnerabilities detection. The dynamic detection process contains five steps: crawler, feature construct, attacks simulation, results detection and report generation. Crawling strategy in crawler module and constructing algorithm in feature construct module are key points of this detection process. Finally, according to the detection technique proposed in this paper, a detection tool is accomplished in Linux using python language to detect web applications. Experiments were launched to verify the results and compare with the test results of other existing tools, analyze the usability, advantages and disadvantages of the detection method above, confirm the feasibility of applying dynamic detection technique to cross-site scripting vulnerabilities detection.

Supervisory control and data acquisition system is an important part of the country's critical infrastructure, but its inherent network characteristics are vulnerable to attack by intruders. The vulnerability of supervisory control and data acquisition system was analyzed, combining common attacks such as information scanning, response injection, command injection and denial of service in industrial control systems, and proposed an intrusion detection model based on graphical features. The time series of message transmission were visualized, extracting the vertex coordinates and various graphic area features to constitute a new data set, and obtained classification model of intrusion detection through training. An intrusion detection experiment environment was built using tools such as MATLAB and power protocol testers. IEC 60870-5-104 protocol which is widely used in power systems had been taken as an example. The results of tests have good effectiveness.

Phishing as one of the most well-known cybercrime activities is a deception of online users to steal their personal or confidential information by impersonating a legitimate website. Several machine learning-based strategies have been proposed to detect phishing websites. These techniques are dependent on the features extracted from the website samples. However, few studies have actually considered efficient feature selection for detecting phishing attacks. In this work, we investigate an agreement on the definitive features which should be used in phishing detection. We apply Fuzzy Rough Set (FRS) theory as a tool to select most effective features from three benchmarked data sets. The selected features are fed into three often used classifiers for phishing detection. To evaluate the FRS feature selection in developing a generalizable phishing detection, the classifiers are trained by a separate out-of-sample data set of 14,000 website samples. The maximum F-measure gained by FRS feature selection is 95% using Random Forest classification. Also, there are 9 universal features selected by FRS over all the three data sets. The F-measure value using this universal feature set is approximately 93% which is a comparable result in contrast to the FRS performance. Since the universal feature set contains no features from third-part services, this finding implies that with no inquiry from external sources, we can gain a faster phishing detection which is also robust toward zero-day attacks.

Phishing is a security attack to acquire personal information like passwords, credit card details or other account details of a user by means of websites or emails. Phishing websites look similar to the legitimate ones which make it difficult for a layman to differentiate between them. As per the reports of Anti Phishing Working Group (APWG) published in December 2018, phishing against banking services and payment processor was high. Almost all the phishy URLs use HTTPS and use redirects to avoid getting detected. This paper presents a focused literature survey of methods available to detect phishing websites. A comparative study of the in-use anti-phishing tools was accomplished and their limitations were acknowledged. We analyzed the URL-based features used in the past to improve their definitions as per the current scenario which is our major contribution. Also, a step wise procedure of designing an anti-phishing model is discussed to construct an efficient framework which adds to our contribution. Observations made out of this study are stated along with recommendations on existing systems.

Peer-to-Peer botnets have become one of the significant threat against network security due to their distributed properties. The decentralized nature makes their detection challenging. It is important to take measures to detect bots as soon as possible to minimize their harm. In this paper, we propose PeerGrep, a novel system capable of identifying P2P bots. PeerGrep starts from identifying hosts that are likely engaged in P2P communications, and then distinguishes P2P bots from P2P hosts by analyzing their active ratio, packet size and the periodicity of connection to destination IP addresses. The evaluation shows that PeerGrep can identify all P2P bots with quite low FPR even if the malicious P2P application and benign P2P application coexist within the same host or there is only one bot in the monitored network.

Brain Computer Interface (BCI) aims at providing a better quality of life to people suffering from neuromuscular disability. This paper establishes a BCI paradigm to provide a biometric security option, used for locking and unlocking personal computers or mobile phones. Although it is primarily meant for the people with neurological disorder, its application can safely be extended for the use of normal people. The proposed scheme decodes the electroencephalogram signals liberated by the brain of the subjects, when they are engaged in selecting a sequence of dots in(6×6)2-dimensional array, representing a pattern lock. The subject, while selecting the right dot in a row, would yield a P300 signal, which is decoded later by the brain-computer interface system to understand the subject's intention. In case the right dots in all the 6 rows are correctly selected, the subject would yield P300 signals six times, which on being decoded by a BCI system would allow the subject to access the system. Because of intra-subjective variation in the amplitude and wave-shape of the P300 signal, a type 2 fuzzy classifier has been employed to classify the presence/absence of the P300 signal in the desired window. A comparison of performances of the proposed classifier with others is also included. The functionality of the proposed system has been validated using the training instances generated for 30 subjects. Experimental results confirm that the classification accuracy for the present scheme is above 90% irrespective of subjects.

As a modern power transmission network, smart grid connects plenty of terminal devices. However, along with the growth of devices are the security threats. Different from the previous separated environment, an adversary nowadays can destroy the power system by attacking these devices. Therefore, it's critical to ensure the security and safety of terminal devices. To achieve this goal, detecting the pre-existing vulnerabilities of the device program and enhance the terminal security, are of great importance and necessity. In this paper, we propose a novel approach that detects existing buffer-overflow vulnerabilities of terminal devices via automatic static analysis (ASA). We utilize the static analysis to extract the device program information and build corresponding program models. By further matching the generated program model with pre-defined vulnerability patterns, we achieve vulnerability detection and error reporting. The evaluation results demonstrate that our method can effectively detect buffer-overflow vulnerabilities of smart terminals with a high accuracy and a low false positive rate.

Routers are important devices in the networks that carry the burden of transmitting information among the communication devices on the Internet. If a malicious adversary wants to intercept the information or paralyze the network, it can directly attack the routers and then achieve the suspicious goals. Thus, preventing router security is of great importance. However, router systems are notoriously difficult to understand or diagnose for their inaccessibility and heterogeneity. The common way of gaining access to the router system and detecting the anomaly behaviors is to inspect the router syslogs or monitor the packets of information flowing to the routers. These approaches just diagnose the routers from one aspect but do not consider them from multiple views. In this paper, we propose an approach to detect the anomalies and faults of the routers with multiple information learning. We try to use the routers' information not from the developer's view but from the user' s view, which does not need any expert knowledge. First, we do the offline learning to transform the benign or corrupted user actions into the syslogs. Then, we try to decide whether the input routers' conditions are poor or not with clustering. During the detection phase, we use the distance between the event and the cluster to decide if it is the anomaly event and we can provide the corresponding solutions. We have applied our approach in a university network which contains Cisco, Huawei and Dlink routers for three months. We aligned our experiment with former work as a baseline for comparison. Our approach can gain 89.6% accuracy in detecting the attacks which is 5.1% higher than the former work. The results show that our approach performs in limited time as well as memory usages and has high detection and low false positives.

According to recent reports, the most common insider threats to systems are unauthorized access to or use of corporate information and exposure of sensitive data. While anomaly detection techniques have proved to be effective in the detection of early signs of data theft, these techniques are not able to detect sophisticated data misuse scenarios in which malicious insiders seek to aggregate knowledge by executing and combining the results of several queries. We thus need techniques that are able to track users' actions across time to detect correlated ones that collectively flag anomalies. In this paper, we propose such techniques for the detection of anomalous accesses to relational databases. Our approach is to monitor users' queries, sequences of queries and sessions of database connection to detect queries that retrieve amounts of data larger than the normal. Our evaluation of the proposed techniques indicates that they are very effective in the detection of anomalies.

Recommender systems try to predict the preferences of users for specific items. These systems suffer from profile injection attacks, where the attackers have some prior knowledge of the system ratings and their goal is to promote or demote a particular item introducing abnormal (anomalous) ratings. The detection of both cases is a challenging problem. In this paper, we propose a framework to spot anomalous rating profiles (outliers), where the outliers hurriedly create a profile that injects into the system either random ratings or specific ratings, without any prior knowledge of the existing ratings. The proposed detection method is based on the unpredictable behavior of the outliers in a validation set, on the user-item rating matrix and on the similarity between users. The proposed system is totally unsupervised, and in the last step it uses the k-means clustering method automatically spotting the spurious profiles. For the cases where labeling sample data is available, a random forest classifier is trained to show how supervised methods outperforms unsupervised ones. Experimental results on the MovieLens 100k and the MovieLens 1M datasets demonstrate the high performance of the proposed schemata.

Collaborative filtering (CF) recommender system has been widely used for its well performing in personalized recommendation, but CF recommender system is vulnerable to shilling attacks in which shilling attack profiles are injected into the system by attackers to affect recommendations. Design robust recommender system and propose attack detection methods are the main research direction to handle shilling attacks, among which unsupervised PCA is particularly effective in experiment, but if we have no information about the number of shilling attack profiles, the unsupervised PCA will be suffered. In this paper, a new unsupervised detection method which combine PCA and data complexity has been proposed to detect shilling attacks. In the proposed method, PCA is used to select suspected attack profiles, and data complexity is used to pick out the authentic profiles from suspected attack profiles. Compared with the traditional PCA, the proposed method could perform well and there is no need to determine the number of shilling attack profiles in advance.

Being able to describe a specific network as consistent is a large step towards resiliency. Next to the importance of security lies the necessity of consistency verification. Attackers are currently focusing on targeting small and crutial goals such as network configurations or flow tables. These types of attacks would defy the whole purpose of a security system when built on top of an inconsistent network. Advances in Artificial Intelligence (AI) are playing a key role in ensuring a fast responce to the large number of evolving threats. Software Defined Networking (SDN), being centralized by design, offers a global overview of the network. Robustness and adaptability are part of a package offered by programmable networking, which drove us to consider the integration between both AI and SDN. The general goal of our series is to achieve an Artificial Intelligence Resiliency System (ARS). The aim of this paper is to propose a new AI-based consistency verification system, which will be part of ARS in our future work. The comparison of different deep learning architectures shows that Convolutional Neural Networks (CNN) give the best results with an accuracy of 99.39% on our dataset and 96% on our consistency test scenario.

Information security has become a growing concern. Computer covert channel which is regarded as an important area of information security research gets more attention. In order to detect these covert channels, a variety of detection algorithms are proposed in the course of the research. The algorithms of machine learning type show better results in these detection algorithms. However, the common machine learning algorithms have many problems in the testing process and have great limitations. Based on the deep learning algorithm, this paper proposes a new idea of network covert channel detection and forms a new detection model. On the one hand, this algorithmic model can detect more complex covert channels and, on the other hand, greatly improve the accuracy of detection due to the use of a new deep learning model. By optimizing this test model, we can get better results on the evaluation index.

A covert channel is a communication channel that is subjugated for illegal flow of information in a way that violates system security policies. It is a dangerous, invisible, undetectable, and developed security attack. Recently, Packet length covert channel has motivated many researchers as it is a one of the most undetectable network covert channels. Packet length covert channel generates a covert traffic that is very similar to normal terrific which complicates the detection of such type of covert channels. This motivates us to introduce a machine learning based detection scheme. Recently, a machine learning approach has proved its capability in many different fields especially in security field as it usually brings up a reliable and realistic results. Based in our developed content and frequency-based features, the developed detection scheme has been fully trained and tested. Our detection scheme has gained an excellent degree of detection accuracy which reaches 98% (zero false negative rate and 0.02 false positive rate).