Biblio

A fundamental recurring task in many machine learning applications is the search for the Nearest Neighbor in high dimensional metric spaces. Towards answering queries in large scale problems, state-of-the-art methods employ Approximate Nearest Neighbors (ANN) search, a search that returns the nearest neighbor with high probability, as well as techniques that compress the dataset. Product-Quantization (PQ) based ANN search methods have demonstrated state-of-the-art performance in several problems, including classification, regression and information retrieval. The dataset is encoded into a Cartesian product of multiple low-dimensional codebooks, enabling faster search and higher compression. Being intrinsically parallel, PQ-based ANN search approaches are amendable for hardware acceleration. This paper proposes a novel Hierarchical PQ (HPQ) based ANN search method as well as an FPGA-tailored architecture for its implementation that outperforms current state of the art systems. HPQ gradually refines the search space, reducing the number of data compares and enabling a pipelined search. The mapping of the architecture on a Stratix 10 FPGA device demonstrates over ×250 speedups over current state-of-the-art systems, opening the space for addressing larger datasets and/or improving the query times of current systems.

Natural language processing (NLP) have been recently used to extract clinical information from free text in Electronic Health Record (EHR). In clinical NLP one challenge is that the meaning of clinical entities is heavily affected by assertion modifiers such as negation, uncertain, hypothetical, experiencer and so on. Incorrect assertion assignment could cause inaccurate diagnosis of patients' condition or negatively influence following study like disease modeling. Thus, clinical NLP systems which can detect assertion status of given target medical findings (e.g. disease, symptom) in clinical context are highly demanded. Here in this work, we propose a deep-learning system based on word embedding, RNN and attention mechanism (more specifically: Attention-based Bidirectional Long Short-Term Memory networks) for assertion detection in clinical notes. Unlike previous state-of-art methods which require knowledge input or feature engineering, our system is a knowledge poor machine learning system and can be easily extended or transferred to other domains. The evaluation of our system on public benchmarking corpora demonstrates that a knowledge poor deep-learning system can also achieve high performance for detecting negation and assertions comparing to state-of-the-art systems.

Once the requirement is gathered in agile, it is broken down into smaller pre-defined format called user stories. These user stories are then scoped in various sprint releases and delivered accordingly. Release planning in Agile becomes challenging when the number of user stories goes up in hundreds. In such scenarios it is very difficult to manually identify similar user stories and package them together into a release. Hence, this paper suggests application of natural language processing algorithms for identifying similar user stories and then scoping them into a release This paper takes the approach to build a word corpus for every project release identified in the project and then to convert the provided user stories into a vector of string using Java utility for calculating top 3 most occurring words from the given project corpus in a user story. Once all the user stories are represented as vector array then by using RV coefficient NLP algorithm the user stories are clustered into various releases of the software project. Using the proposed approach, the release planning for large and complex software engineering projects can be simplified resulting into efficient planning in less time. The automated commercial tools like JIRA and Rally can be enhanced to include suggested algorithms for managing release planning in Agile.

With the rapid growth of mobile devices and mobile apps, mobile has surpassed desktop and now has the largest worldwide market share [1]. While such growth brings in more opportunities, it also poses new challenges in security. Among the challenges, user privacy protection has drawn tremendous attention in recent years, especially after the Facebook-Cambridge Analytica data scandal in April 2018 [2].

Natural language processing is the core of machine translation. In the history, its development process is almost the same as machine translation, and the two complement each other. This article compares the natural language processing of statistical corpora with neural machine translation and concludes the natural language processing: Neural machine translation has the advantage of deep learning, which is very suitable for dealing with the high dimension, label-free and big data of natural language, therefore, its application is more general and reflects the power of big data and big data thinking.

In the Network on Chip (NoC), performance optimization has always been a research focus. Compared with the static routing scheme, dynamical routing schemes can better reduce the data of packet transmission latency under network congestion. In this paper, we propose a dynamical Q-learning routing approach with real-time monitoring of NoC. Firstly, we design a real-time monitoring scheme and the corresponding circuits to record the status of traffic congestion for NoC. Secondly, we propose a novel method of Q-learning. This method finds an optimal path based on the lowest traffic congestion. Finally, we dynamically redistribute network tasks to increase the packet transmission speed and balance the traffic load. Compared with the C-XY routing and DyXY routing, our method achieved improvement in terms of 25.6%-49.5% and 22.9%-43.8%.

With increasing integration in SoCs, the Network-on-Chip (NoC) connecting cores and accelerators is of paramount importance to provide low-latency and high-throughput communication. Due to limits to scaling of electrical wires in terms of energy and delay, especially for long multi-mm distances on-chip, alternate technologies such as Wireless Network-on-Chip (WNoC) have shown promise. WNoCs can provide low-latency one-hop broadcasts across the entire chip and can augment point-to-point multi-hop signaling over traditional wired NoCs. Thus, there has been a recent surge in research demonstrating the performance and energy benefits of WNoCs. However, little to no work has studied the additional security and fault tolerance challenges that are unique to WNoCs. In this work, we study potential threats related to denial-of-service, spoofing, and eavesdropping attacks in WNoCs, due to malicious hardware trojans or faulty wireless components. We introduce Prometheus, a dropin solution inside the network interface that provides protection from all three attacks, while adhering to the strict area, power and latency constraints of on-chip systems.

To gain strategic insight into defending against the network reconnaissance stage of advanced persistent threats, we recreate the escalating competition between scans and deceptive views on a Software Defined Network (SDN). Our threat model presumes the defense is a deceptive network view unique for each node on the network. It can be configured in terms of the number of honeypots and subnets, as well as how real nodes are distributed across the subnets. It assumes attacks are NMAP ping scans that can be configured in terms of how many IP addresses are scanned and how they are visited. Higher performing defenses detect the scanner quicker while leaking as little information as possible while higher performing attacks are better at evading detection and discovering real nodes. By using Artificial Intelligence in the form of a competitive coevolutionary genetic algorithm, we can analyze the configurations of high performing static defenses and attacks versus their evolving adversary as well as the optimized configuration of the adversary itself. When attacks and defenses both evolve, we can observe that the extent of evolution influences the best configurations.

An information system based on real-time file integrations has an important role in today's organizations' work process management. By connecting to the network, file flow and integration between corporate systems have gained a great significance. In addition, network and security issues have emerged depending on the file structure and transfer processes. Thus, there has become a need for an effective and self-learning anomaly detection module for file transfer processes in order to provide the persistence of integration channels, accountability of transfer logs and data integrity. This paper proposes a novel anomaly detection approach that focuses on file size and integration duration of file transfers between enterprise systems. For this purpose, size and time anomalies on transferring files will be detected by a machine learning-based structure. Later, an alarm system is going to be developed in order to inform the authenticated individuals about the anomalies.

Anomaly-based Detection Systems (ADSs) attempt to learn the features of behaviors and events of a system and/or users over a period to build a profile of normal behaviors. There has been a growing interest in ADSs and typically conceived as more powerful systems One of the important factors for ADSs is an ability to distinguish between normal and abnormal behaviors in a given period. However, it is getting complicated due to the dynamic network environment that changes every minute. It is dangerous to distinguish between normal and abnormal behaviors with a fixed threshold in a dynamic environment because it cannot guarantee the threshold is always an indication of normal behaviors. In this paper, we propose an adaptive threshold for a dynamic environment with a trust management scheme for efficiently managing the profiles of normal and abnormal behaviors. Based on the assumption of the statistical analysis-based ADS that normal data instances occur in high probability regions while malicious data instances occur in low probability regions of a stochastic model, we set two adaptive thresholds for normal and abnormal behaviors. The behaviors between the two thresholds are classified as suspicious behaviors, and they are efficiently evaluated with a trust management scheme.

One of the latest emerging technologies is artificial intelligence, which makes the machine mimic human behavior. The most important component used to detect cyber attacks or malicious activities is the Intrusion Detection System (IDS). Artificial intelligence plays a vital role in detecting intrusions and widely considered as the better way in adapting and building IDS. In trendy days, artificial intelligence algorithms are rising as a brand new computing technique which will be applied to actual time issues. In modern days, neural network algorithms are emerging as a new artificial intelligence technique that can be applied to real-time problems. The proposed system is to detect a classification of botnet attack which poses a serious threat to financial sectors and banking services. The proposed system is created by applying artificial intelligence on a realistic cyber defense dataset (CSE-CIC-IDS2018), the very latest Intrusion Detection Dataset created in 2018 by Canadian Institute for Cybersecurity (CIC) on AWS (Amazon Web Services). The proposed system of Artificial Neural Networks provides an outstanding performance of Accuracy score is 99.97% and an average area under ROC (Receiver Operator Characteristic) curve is 0.999 and an average False Positive rate is a mere value of 0.001. The proposed system using artificial intelligence of botnet attack detection is powerful, more accurate and precise. The novel proposed system can be implemented in n machines to conventional network traffic analysis, cyber-physical system traffic data and also to the real-time network traffic analysis.

Conventional methods for anomaly detection include techniques based on clustering, proximity or classification. With the rapidly growing social networks, outliers or anomalies find ingenious ways to obscure themselves in the network and making the conventional techniques inefficient. In this paper, we utilize the ability of Deep Learning over topological characteristics of a social network to detect anomalies in email network and twitter network. We present a model, Graph Neural Network, which is applied on social connection graphs to detect anomalies. The combinations of various social network statistical measures are taken into account to study the graph structure and functioning of the anomalous nodes by employing deep neural networks on it. The hidden layer of the neural network plays an important role in finding the impact of statistical measure combination in anomaly detection.

In this paper, it first analyzed the characteristics of Advanced Persistent Threat (APT). according to APT attack, this paper established an BP neural network optimized by improved adaptive genetic algorithm to predict the security risk of nodes in the network. and calculated the path of APT attacks with the maximum possible attack. Finally, experiments verify the effectiveness and correctness of the algorithm by simulating attacks. Experiments show that this model can effectively evaluate the security situation in the network, For the defenders to adopt effective measures defend against APT attacks, thus improving the security of the network.

The growing complexity and diversification of cyber-attacks are largely reflected in the increasing sophistication of security appliances, which are often too cumbersome to be run in virtual services and IoT devices. Hence, the design of cyber-security frameworks is today looking at more cooperative models, which collect security-related data from a large set of heterogeneous sources for centralized analysis and correlation.In this paper, we outline a flexible abstraction layer for access to security context. It is conceived to program and gather data from lightweight inspection and enforcement hooks deployed in cloud applications and IoT devices. We also provide a preliminary description of its implementation, by reviewing the main software components and their role.

Network architecture design and verification has become increasingly complicated as a greater number of security considerations, implementations, and factors are included in the design process. In the design process, one must account for various costs of interwoven layers of security. Generally these costs are simplified for evaluation of risk to the network. The obvious implications of adding security are the need to account for the impacts of loss (risk) and accounting for the ensuing increased design costs. The considerations that are not traditionally examined are those of the adversary and the defender of a given system. Without accounting for the view point of the individuals interacting with a network architecture, one can not verify and select the most advantageous security implementation. This work presents a method for obtaining a security metric that takes into account not only the risk of the defender, but also the probability of an attack originating from the motivation of the adversary. We then move to a more meaningful metric based on a monetary unit that architects can use in choosing a best fit solution for a given network critical path design problem.

As power grid control systems become increasingly automated and distributed, security has become a significant design concern. Systems increasingly expose new avenues, at a variety of levels, for attackers to exploit and enable widespread disruptions and/or surveillance. Much prior work has explored the implications of attack models focused on false data injection at the front-end of the control system (i.e. during state estimation) [1]. Instead, in this paper we focus on characterizing the inherent cyber-attack vulnerabilities with power flow. Power flow (and power flow constraints) are at the core of many applications critical to operation of power grids (e.g. state estimation, economic dispatch, contingency analysis, etc.). We propose two algorithmic approaches for characterizing the vulnerability of buses within power grids to cyber-attacks. Specifically, we focus on measuring the instability of power flow to attacks which manifest as either voltage or power related errors. Our results show that attacks manifesting as voltage errors are an order of magnitude more likely to cause instability than attacks manifesting as power related errors (and 5x more likely for state estimation as compared to power flow).

The growth of the internet has brought along positive gains such as the emergence of a highly interconnected world. However, on the flip side, there has been a growing concern on how secure distributed systems can be built effectively and tested for security vulnerabilities prior to deployment. Developing a secure software product calls for a deep technical understanding of some complex issues with regards to the software and its operating environment, as well as embracing a systematic approach of analyzing the software. This paper proposes a method for identifying software security vulnerabilities from software requirement specifications written in Structured Object-oriented Formal Language (SOFL). Our proposed methodology leverages on the concept of providing an early focus on security by identifying potential security vulnerabilities at the requirement analysis and verification phase of the software development life cycle.

Wide bandgap power semiconductors are key enablers for increasing the power density of switch-mode power supplies. However, they require new gate drive technologies. This paper examines and characterizes a fabricated gate-driver in a class-E resonant inverter. The gate-driver's total area of 1.2mm2 includes two high-voltage transistors for gate-driving, integrated complementary metal-oxide-semiconductor (CMOS) gate-drivers, high-speed floating level-shifter and reset circuitry. A prototype printed circuit board (PCB) was designed to assess the implications of an electrostatic discharge (ESD) diode, its parasitic capacitance and package bondwire connections. The parasitic capacitance was estimated using its discharge time from an initial voltage and the capacitance is 56.7 pF. Both bondwires and the diode's parasitic capacitance is neglegible. The gate-driver's functional behaviour is validated using a parallel LC resonant tank resembling a self-oscillating gate-drive. Measurements and simulations show the ESD diode clamps the output voltage to a minimum of -2V.

Mutriku wave farm is the first commercial plant all around the world. Since July 2011 it has been continuously selling electricity to the grid. It operates with the OWC technology and has 14 operating Wells-type turbines. In the plant there is a SCADA data recording system that collects the most important parameters of the turbines; among them, the pressure in the inlet chamber, the position of the security valve (from fully open to fully closed) and the generated power in the last 5 minutes. There is also an electricity meter which provides information about the amount of electric energy sold to the grid. The 2014 winter (January, February and March), and especially the first fortnight of February, was a stormy winter with rough sea state conditions. This was reflected both in the performance of the turbines (high pressure values, up to 9234.2 Pa; low opening degrees of the security valve, down to 49.4°; and high power generation of about 7681.6 W, all these data being average values) and in the calculated capacity factor (CF = 0.265 in winter and CF = 0.294 in February 2014). This capacity factor is a good tool for the comparison of different WEC technologies or different locations and shows an important seasonal behavior.

ARGOS is a web service we implemented to offer face recognition Authentication Services (AaaS) to mobile and desktop (via the web browser) end users. The Authentication Services may be used by 3rd party service organizations to enhance their service offering to their customers. ARGOS implements a secure face recognition-based authentication service aiming to provide simple and intuitive tools for 3rd party service providers (like PayPal, banks, e-commerce etc) to replace passwords with face biometrics. It supports authentication from any device with 2D or 3D frontal facing camera (mobile phones, laptops, tablets etc.) and almost any operating systems (iOS, Android, Windows and Linux Ubuntu).

Bitcoin is popular not only with consumers, but also with cybercriminals (e.g., in ransomware and online extortion, and commercial online child exploitation). Given the potential of Bitcoin to be involved in a criminal investigation, the need to have an up-to-date and in-depth understanding on the forensic acquisition and analysis of Bitcoins is crucial. However, there has been limited forensic research of Bitcoin in the literature. The general focus of existing research is on postmortem analysis of specific locations (e.g. wallets on mobile devices), rather than a forensic approach that combines live data forensics and postmortem analysis to facilitate the identification, acquisition, and analysis of forensic traces relating to the use of Bitcoins on a system. Hence, the latter is the focus of this paper where we present an open source tool for live forensic and postmortem analysing automatically. Using this open source tool, we describe a list of target artifacts that can be obtained from a forensic investigation of popular Bitcoin clients and Web Wallets on different web browsers installed on Windows 7 and Windows 10 platforms.

With the remarkable success of deep learning, Deep Neural Networks (DNNs) have been applied as dominant tools to various machine learning domains. Despite this success, however, it has been found that DNNs are surprisingly vulnerable to malicious attacks; adding a small, perceptually indistinguishable perturbations to the data can easily degrade classification performance. Adversarial training is an effective defense strategy to train a robust classifier. In this work, we propose to utilize the generator to learn how to create adversarial examples. Unlike the existing approaches that create a one-shot perturbation by a deterministic generator, we propose a recursive and stochastic generator that produces much stronger and diverse perturbations that comprehensively reveal the vulnerability of the target classifier. Our experiment results on MNIST and CIFAR-10 datasets show that the classifier adversarially trained with our method yields more robust performance over various white-box and black-box attacks.

HTTPS provides authentication, data confidentiality, and integrity for secure web applications in the Internet. In order to establish secure connections with the target website but not a man-in-the-middle or impersonation attacker, a browser shows security warnings to users, when different HTTPS errors happen (e.g., it fails to build a valid certificate chain, or the certificate subject does not match the domain visited). Each browser implements its own design of warnings on HTTPS errors, to balance security and usability. This paper presents a list of common HTTPS errors, and we investigate the browser behaviors on each error. Our study discloses browser defects on handling HTTPS errors in terms of cryptographic algorithm, certificate verification, name validation, HPKP, and HSTS.

Every public or private area today is preferred to be under surveillance to ensure high levels of security. Since the surveillance happens round the clock, data gathered as a result is huge and requires a lot of manual work to go through every second of the recorded videos. This paper presents a system which can detect anomalous behaviors and alarm the user on the type of anomalous behavior. Since there are a myriad of anomalies, the classification of anomalies had to be narrowed down. There are certain anomalies which are generally seen and have a huge impact on public safety, such as explosions, road accidents, assault, shooting, etc. To narrow down the variations, this system can detect explosion, road accidents, shooting, and fighting and even output the frame of their occurrence. The model has been trained with videos belonging to these classes. The dataset used is UCF Crime dataset. Learning patterns from videos requires the learning of both spatial and temporal features. Convolutional Neural Networks (CNN) extract spatial features and Long Short-Term Memory (LSTM) networks learn the sequences. The classification, using an CNN-LSTM model achieves an accuracy of 85%.

Recently, Pan/Tilt/Zoom (PTZ) camera has been widely used in video surveillance systems. However, it is difficult to automatically control PTZ cameras according to moving objects in the surveillance area. This paper proposes an automatic camera control method based on a Deep-Q Network (DQN) for improving the recognition accuracy of anomaly actions in the video surveillance system. To generate PTZ camera control values, the proposed method uses the position and size information of the object which received from the video analysis system. Through implementation results, the proposed method can automatically control the PTZ camera according to moving objects.