Biblio

Machine learning experts prefer to think of their input as a single, homogeneous, and consistent data set. However, when analyzing large volumes of data, the entire data set may not be manageable on a single server, but must be stored on a distributed file system instead. Moreover, with the pressing demand to deliver explainable models, the experts may no longer focus on the machine learning algorithms in isolation, but must take into account the distributed nature of the data stored, as well as the impact of any data pre-processing steps upstream in their data analysis pipeline. In this paper, we make the point that even basic transformations during data preparation can impact the model learned, and that this is exacerbated in a distributed setting. We then sketch our vision of end-to-end explainability of the model learned, taking the pre-processing into account. In particular, we point out the potentials of linking the contributions of research on data provenance with the efforts on explainability in machine learning. In doing so, we highlight pitfalls we may experience in a distributed system on the way to generating more holistic explanations for our machine learning models.

Machine Learning (ML) has become essential in several industries. In Computational Science and Engineering (CSE), the complexity of the ML lifecycle comes from the large variety of data, scientists' expertise, tools, and workflows. If data are not tracked properly during the lifecycle, it becomes unfeasible to recreate a ML model from scratch or to explain to stackholders how it was created. The main limitation of provenance tracking solutions is that they cannot cope with provenance capture and integration of domain and ML data processed in the multiple workflows in the lifecycle, while keeping the provenance capture overhead low. To handle this problem, in this paper we contribute with a detailed characterization of provenance data in the ML lifecycle in CSE; a new provenance data representation, called PROV-ML, built on top of W3C PROV and ML Schema; and extensions to a system that tracks provenance from multiple workflows to address the characteristics of ML and CSE, and to allow for provenance queries with a standard vocabulary. We show a practical use in a real case in the O&G industry, along with its evaluation using 239,616 CUDA cores in parallel.

Explaining reasoning and behaviour of artificial intelligent systems to human users becomes increasingly urgent, especially in the field of machine learning. Many recent contributions approach this issue with post-hoc methods, meaning they consider the final system and its outcomes, while the roots of included artefacts are widely neglected. However, we argue in this position paper that there needs to be a stronger focus on the development process. Without insights into specific design decisions and meta information that accrue during the development an accurate explanation of the resulting model is hardly possible. To remedy this situation we propose to increase process transparency by applying provenance methods, which serves also as a basis for increased explainability.

With the developments in deep learning, the security of neural networks against vulnerabilities has become one of the most urgent research topics in deep learning. There are many types of security countermeasures. Adversarial examples and their defense methods, in particular, have been well-studied in recent years. An adversarial example is designed to make neural networks misclassify or produce inaccurate output. Audio adversarial examples are a type of adversarial example where the main target of attack is a speech-to-text transcription neural network. In this study, we propose a new defense method against audio adversarial examples for the speech-to-text transcription neural networks. It is difficult to determine whether an input waveform data representing the sound of voice is an audio adversarial example. Therefore, the main framework of the proposed defense method is based on a sandbox approach. To evaluate the proposed defense method, we used actual audio adversarial examples that were created on Deep Speech, which is a speech-to-text transcription neural network. We confirmed that our defense method can identify audio adversarial examples to protect speech-to-text systems.

With the development of technology, the massive malware become the major challenge to current computer security. In our work, we implemented a malware detection system using deep learning on API calls. By means of cuckoo sandbox, we extracted the API calls sequence of malicious programs. Through filtering and ordering the redundant API calls, we extracted the valid API sequences. Compared with GRU, BGRU, LSTM and SimpleRNN, we evaluated the BLSTM on the massive datasets including 21,378 samples. The experimental results demonstrate that BLSTM has the best performance for malware detection, reaching the accuracy of 97.85%.

For the task with complicated manipulation in unstructured environments, traditional hand-coded methods are ineffective, while reinforcement learning can provide more general and useful policy. Although the reinforcement learning is able to obtain impressive results, its stability and reliability is hard to guarantee, which would cause the potential safety threats. Besides, the transfer from simulation to real-world also will lead in unpredictable situations. To enhance the safety and reliability of robots, we introduce the force and haptic perception into reinforcement learning. Force and tactual sensation play key roles in robotic dynamic control and human-robot interaction. We demonstrate that the force-based reinforcement learning method can be more adaptive to environment, especially in sim-to-real transfer. Experimental results show in object pushing task, our strategy is safer and more efficient in both simulation and real world, thus it holds prospects for a wide variety of robotic applications.

With the rapid increase in the number of Internet of Things (IoT) devices, mobile devices, cloud services, and cyber-physical systems, the large-scale cyber attacks on enterprises and public sectors have increased. In particular, ransomware attacks damaged UK's National Health Service and many enterprises around the world in 2017. Therefore, researchers have proposed ransomware detection and prevention systems. However, manual inspection in static and dynamic ransomware analysis is time-consuming and it cannot cope with the rapid increase in variants of ransomware family. Recently, machine learning has been used to automate ransomware analysis by creating a behavioral model of same ransomware family. To create effective behavioral models of ransomware, we first obtained storage access patterns of live ransomware samples and of a benign application by using a live-forensic hypervisor called WaybackVisor. To distinguish ransomware from a benign application that has similar behavior to ransomware, we carefully selected five dimensional features that were extracted both from actual ransomware's Input and Output (I/O) logs and from a benign program's I/O logs. We created and evaluated machine learning models by using Random Forest, Support Vector Machine, and K-Nearest Neighbors. Our experiments using the proposed five features of storage access patterns achieved F-measure rate of 98%.

In parallel with the increasing growth of the Internet and computer networks, the number of malwares has been increasing every day. Today, one of the newest attacks and the biggest threats in cybersecurity is ransomware. The effectiveness of applying machine learning techniques for malware detection has been explored in much scientific research, however, there is few studies focused on machine learning-based ransomware detection. In this paper, the effectiveness of ransomware detection using machine learning methods applied to CICAndMal2017 dataset is examined in two experiments. First, the classifiers are trained on a single dataset containing different types of ransomware. Second, different classifiers are trained on datasets of 10 ransomware families distinctly. Our findings imply that in both experiments random forest outperforms other tested classifiers and the performance of the classifiers are not changed significantly when they are trained on each family distinctly. Therefore, the random forest classification method is very effective in ransomware detection.

With the increasing diversity of Android malware, the effectiveness of conventional defense mechanisms are at risk. This situation has endorsed a notable interest in the improvement of the exactitude and scalability of malware detection for smart devices. In this study, we have proposed an effective deep learning-based malware detection model for competent and improved ransomware detection in Android environment by looking at the algorithm of Long Short-Term Memory (LSTM). The feature selection has been done using 8 different feature selection algorithms. The 19 important features are selected through simple majority voting process by comparing results of all feature filtration techniques. The proposed algorithm is evaluated using android malware dataset (CI-CAndMal2017) and standard performance parameters. The proposed model outperforms with 97.08% detection accuracy. Based on outstanding performance, we endorse our proposed algorithm to be efficient in malware and forensic analysis.

With the development of the Internet, the network attack technology has undergone tremendous changes. The forms of network attack and defense have also changed, which are features in attacks are becoming more diverse, attacks are more widespread and traditional security protection methods are invalid. In recent years, with the development of software defined security, network anomaly detection technology and big data technology, these challenges have been effectively addressed. This paper proposes a data-driven software defined security architecture with core features including data-driven orchestration engine, scalable network anomaly detection module and security data platform. Based on the construction of the analysis layer in the security data platform, real-time online detection of network data can be realized by integrating network anomaly detection module and security data platform under software defined security architecture. Then, data-driven security business orchestration can be realized to achieve efficient, real-time and dynamic response to detected anomalies. Meanwhile, this paper designs a deep learning-based HTTP anomaly detection algorithm module and integrates it with data-driven software defined security architecture so that demonstrating the flow of the whole system.

Supervisory Control and Data Acquisition (SCADA)networks are widely deployed in modern industrial control systems (ICSs)such as energy-delivery systems. As an increasing number of field devices and computing nodes get interconnected, network-based cyber attacks have become major cyber threats to ICS network infrastructure. Field devices and computing nodes in ICSs are subjected to both conventional network attacks and specialized attacks purposely crafted for SCADA network protocols. In this paper, we propose a deep-learning-based network intrusion detection system for SCADA networks to protect ICSs from both conventional and SCADA specific network-based attacks. Instead of relying on hand-crafted features for individual network packets or flows, our proposed approach employs a convolutional neural network (CNN)to characterize salient temporal patterns of SCADA traffic and identify time windows where network attacks are present. In addition, we design a re-training scheme to handle previously unseen network attack instances, enabling SCADA system operators to extend our neural network models with site-specific network attack traces. Our results using realistic SCADA traffic data sets show that the proposed deep-learning-based approach is well-suited for network intrusion detection in SCADA systems, achieving high detection accuracy and providing the capability to handle newly emerged threats.

The continuing decrease in feature size of integrated circuits, and the increase of the complexity and cost of design and fabrication has led to outsourcing the design and fabrication of integrated circuits to third parties across the globe, and in turn has introduced several security vulnerabilities. The adversaries in the supply chain can pirate integrated circuits, overproduce these circuits, perform reverse engineering, and/or insert hardware Trojans in these circuits. Developing countermeasures against such security threats is highly crucial. Accordingly, this paper first develops a learning-based trust verification framework to detect hardware Trojans. To tackle Trojan insertion, IP piracy and overproduction, logic locking schemes and in particular stripped functionality logic locking is discussed and its resiliency against the state-of-the-art attacks is investigated.

Fingerprinting the malware by its behavioural signature has been an attractive approach for malware detection due to the homogeneity of dynamic execution patterns across different variants of similar families. Although previous researches show reasonably good performance in dynamic detection using machine learning techniques on a large corpus of training set, decisions must be undertaken based upon a scarce number of observable samples in many practical defence scenarios. This paper demonstrates the effectiveness of generative adversarial autoencoder for dynamic malware detection under outbreak situations where in most cases a single sample is available for training the machine learning algorithm to detect similar samples that are in the wild.

Logic locking, and Integrated Circuit (IC) Camouflaging, are techniques that try to hide the design of an IC from a malicious foundry or end-user by introducing ambiguity into the netlist of the circuit. While over the past decade an array of such techniques have been proposed, their security has been constantly challenged by algorithmic attacks. This may in part be due to a lack of formally defined notions of security in the first place, and hence a lack of security guarantees based on long-standing hardness assumptions. In this paper we take a formal approach. We define the problem of circuit locking (cL) as transforming an original circuit to a locked one which is ``unintelligable'' without a secret key (this can model camouflaging and split-manufacturing in addition to logic locking). We define several notions of security for cL under different adversary models. Using long standing results from computational learning theory we show the impossibility of exponentially approximation-resilient locking in the presence of an oracle for large classes of Boolean circuits. We then show how exact-recovery-resiliency and a more relaxed notion of security that we coin ``best-possible'' approximation-resiliency can be provably guaranteed with polynomial overhead. Our theoretical analysis directly results in stronger attacks and defenses which we demonstrate through experimental results on benchmark circuits.

We recently see a real digital revolution where all companies prefer to use cloud computing because of its capability to offer a simplest way to deploy the needed services. However, this digital transformation has generated different security challenges as the privacy vulnerability against cyber-attacks. In this work we will present a new architecture of a hybrid Intrusion detection System, IDS for virtual private clouds, this architecture combines both network-based and host-based intrusion detection system to overcome the limitation of each other, in case the intruder bypassed the Network-based IDS and gained access to a host, in intend to enhance security in private cloud environments. We propose to use a non-traditional mechanism in the conception of the IDS (the detection engine). Machine learning, ML algorithms will can be used to build the IDS in both parts, to detect malicious traffic in the Network-based part as an additional layer for network security, and also detect anomalies in the Host-based part to provide more privacy and confidentiality in the virtual machine. It's not in our scope to train an Artificial Neural Network ”ANN”, but just to propose a new scheme for IDS based ANN, In our future work we will present all the details related to the architecture and parameters of the ANN, as well as the results of some real experiments.

The failure prediction method of virtual machines (VM) guarantees reliability to cloud platforms. However, the uncertainty of VM security state will affect the reliability and task processing capabilities of the entire cloud platform. In this study, a failure prediction method of VM based on AdaBoost-Hidden Markov Model was proposed to improve the reliability of VMs and overall performance of cloud platforms. This method analyzed the deep relationship between the observation state and the hidden state of the VM through the hidden Markov model, proved the influence of the AdaBoost algorithm on the hidden Markov model (HMM), and realized the prediction of the VM failure state. Results show that the proposed method adapts to the complex dynamic cloud platform environment, can effectively predict the failure state of VMs, and improve the predictive ability of VM security state.

Cloud Computing as of large is evolving at a faster pace with an ever changing set of cloud services. The amenities in the cloud are all enabled with respect to the public cloud services in their own enormous domain aspects commercially, which tend to be more insecure. These cloud services should be thus protected and secured which is very vital to the cloud infrastructures. Therefore, in this research work, we have identified security features with a self-heal approach that could be rendered on the infrastructure as a service (IaaS) in a private cloud environment. We have investigated the attack model from the virtual machine snapshots and have analyzed based on the supervised machine learning techniques. The virtual machines memory snapshots API call sequences are considered as input for the supervised and unsupervised machine learning algorithms to classify the attacked and the un-attacked virtual machine memory snapshots. The obtained set of the attacked virtual machine memory snapshots are given as input to the self-heal algorithm which is enabled to retrieve back the functionality of the virtual machines. Our method of detecting the malware attains about 93% of accuracy with respect to the virtual machine snapshots.

We propose a reinforcement learning based online multi-tenant secret-key assignment algorithm for quantum key distribution networks, capable of reducing tenant-request blocking probability more than half compared to the benchmark heuristics.

This study proposes a method for ECG signals quality assessment (SQA) by using temporal feature, and heuristic rule. The ECG signal will be classified as acceptable or unacceptable. Seven types of noise were able to be detected by the prosed method. The noises are: FL, TVN, BW, AB, MA, PLI and AWGN. The proposed method is aimed to have better performance for SQA than classical machine learning method. The experiment is conducted by using 1000 instances ECG signal. The experiment result shows that db8 has the best performance with 0.86, 0.85 and 85.6% on lead-1 signal and 0.69, 0.79, and 74% on lead-5 signal for specificity, sensitivity and accuracy respectively. Compared to the classical machine learning, the proposed heuristic method has same accuracy but has 48% and 31% better specificity for lead-1 and lead-5. It means that the proposed method has far better ability to detect noise.

Cross-Site Request Forgery (CSRF) is one of the oldest and simplest attacks on the Web, yet it is still effective on many websites and it can lead to severe consequences, such as economic losses and account takeovers. Unfortunately, tools and techniques proposed so far to identify CSRF vulnerabilities either need manual reviewing by human experts or assume the availability of the source code of the web application. In this paper we present Mitch, the first machine learning solution for the black-box detection of CSRF vulnerabilities. At the core of Mitch there is an automated detector of sensitive HTTP requests, i.e., requests which require protection against CSRF for security reasons. We trained the detector using supervised learning techniques on a dataset of 5,828 HTTP requests collected on popular websites, which we make available to other security researchers. Our solution outperforms existing detection heuristics proposed in the literature, allowing us to identify 35 new CSRF vulnerabilities on 20 major websites and 3 previously undetected CSRF vulnerabilities on production software already analyzed using a state-of-the-art tool.

Vehicular networks are susceptible to variety of attacks such as denial of service (DoS) attack, sybil attack and false alert generation attack. Different cryptographic methods have been proposed to protect vehicular networks from these kind of attacks. However, cryptographic methods have been found to be less effective to protect from insider attacks which are generated within the vehicular network system. Misbehavior detection system is found to be more effective to detect and prevent insider attacks. In this paper, we propose a machine learning based misbehavior detection system which is trained using datasets generated through extensive simulation based on realistic vehicular network environment. The simulation results demonstrate that our proposed scheme outperforms previous methods in terms of accurately identifying various misbehavior.

Cascading failure, which can be triggered by both physical and cyber attacks, is among the most critical threats to the security and resilience of power grids. In current literature, researchers investigate the issue of cascading failure on smart grids mainly from the attacker's perspective. From the perspective of a grid defender or operator, however, it is also an important issue to restore the smart grid suffering from cascading failure back to normal operation as soon as possible. In this paper, we consider cascading failure in conjunction with the restoration process involving repairing of the failed nodes/links in a sequential fashion. Based on a realistic power flow cascading failure model, we exploit a Q-learning approach to develop a practical and effective policy to identify the optimal way of sequential restorations for large-scale smart grids. Simulation results on three power grid test benchmarks demonstrate the learning ability and the effectiveness of the proposed strategy.

Cyber-Physical Systems (CPS) are growing with added complexity and functionality. Multidisciplinary interactions with physical systems are the major keys to CPS. However, sensors, actuators, controllers, and wireless communications are prone to attacks that compromise the system. Machine learning models have been utilized in controllers of automotive to learn, estimate, and provide the required intelligence in the control process. However, their estimation is also vulnerable to the attacks from physical or cyber domains. They have shown unreliable predictions against unknown biases resulted from the modeling. In this paper, we propose a novel control design using conditional generative adversarial networks that will enable a self-secured controller to capture the normal behavior of the control loop and the physical system, detect the anomaly, and recover from them. We experimented our novel control design on a self-secured BMS by driving a Nissan Leaf S on standard driving cycles while under various attacks. The performance of the design has been compared to the state-of-the-art; the self-secured BMS could detect the attacks with 83% accuracy and the recovery estimation error of 21% on average, which have improved by 28% and 8%, respectively.

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.

Keeping Internet users safe from attacks and other threats is one of the biggest security challenges nowadays. Distributed Denial of Service (DDoS) [1] is one of the most common attacks. DDoS makes the system stop working by resource overload. Software Define Networking (SDN) [2] has recently emerged as a new networking technology offering an unprecedented programmability that allows network operators to dynamically configure and manage their infrastructures. The flexible processing and centralized management of SDN controller allow flexibly deploying complex security algorithms and mitigation methods. In this paper, we propose a new TCP-SYN flood attack mitigation in SDN networks using machine learning. By using a testbed, we implement the proposed algorithms, evaluate their accuracy and address the trade-off between the accuracy and capacity of the security device. The results show that the algorithms can mitigate TCP-SYN Flood attack over 96.