To explore the advantages of adversarial learning and deep learning, we propose a novel network intrusion detection model called SAVAER-DNN, which can not only detect known unknown attacks but also improve rate low-frequent attacks. SAVAER is supervised variational auto-encoder with regularization, uses WGAN-GP instead vanilla GAN to learn latent distribution original data. SAVAER's decoder used synthesize samples attacks, thereby increasing diversity training balancing data set. encoder...

This technical paper presents the method that we use in Road Damage Detection and Classification Challenge, which is designed to detect damages contained road images photographed by a vehicle-mounted smartphone. In this task, apply Faster R-CNN classify damaged roads. Through analyses of aspect ratios sizes areas training dataset, adjust relevant parameters model. order solve problem unbalanced data distribution different classes, introduce some augmentation techniques (contrast...

Phishing has become one of the biggest and most effective cyber threats, causing hundreds millions dollars in losses data breaches every year. Currently, anti-phishing techniques require experts to extract phishing sites features use third-party services detect sites. These have some limitations, which is that extracting requires expertise time-consuming. Second, delays detection Hence, this paper proposes an integrated website method based on convolutional neural networks (CNN) random...

Cloud computing is an Internet-based pattern through which shared resources are provided to devices on-demand. It emerging but promising paradigm integrating mobile into cloud computing, and the integration performs in based hierarchical multi-user data-shared environment. With security issues such as data confidentiality user authority may arise system, it concerned main constraints developments of computing. In order provide safe secure operation, a access control method using modified...

The detection and removal of malicious social bots in networks has become an area interest industry academia. widely used bot method based on machine learning leads to imbalance the number samples different categories. Classifier bias a low rate minority samples. Therefore, we propose improved conditional generative adversarial network (improved CGAN) extend imbalanced data sets before applying training classifiers improve accuracy bots. To generate auxiliary condition, modified clustering...

This paper presents the method that underlies our submission to popularity prediction task of Social Media Prediction Challenge 2017. The is designed predict impact sharing different posts for a publisher on social media. There are many factors influence image popularity; these include not only visual features image, but also features, such as user characteristics its poster and even upload time. In this project, we propose fast effective framework prediction. First, investigate extract...

Nowadays, the Internet is flooded with huge traffic, many applications have millions users, a single server difficult to bear large number of clients' access, so application providers will put several servers as computing unit provide support for specific application, usually people use distributed computing, load balancing technology complete work. A typical technique dedicated balancer forward client requests different servers, this requires hardware support, expensive, lacks flexibility...

Software defined network (SDN) is an emerging architecture. Its control logic and forwarding are separated. SDN has the characteristics of centralized management, which makes it easier for malicious attackers to use security vulnerabilities networks implement distributed denial Service (DDoS) attack. Information entropy a kind lightweight DDoS early detection method. This paper proposes attack method in based on φ-entropy. φ-entropy can adjust related parameters according conditions enlarge...

While antiphishing techniques have evolved over the years, phishing remains one of most threatening attacks on current network security. This is because exploits weakest links in a system—people. The purpose this research to predict possible victims. In study, we propose multidimensional susceptibility prediction model (MPSPM) implement user susceptibility. We constructed two types emails: legitimate emails and emails. gathered 1105 volunteers join our experiment by recruiting volunteers....

We consider traffic scheduling in an N times packet switch with optical fabric, where the fabric requires a reconfiguration overhead to change its configurations. To provide 100% throughput bounded delay, speedup is necessary compensate for both and inefficiency of algorithm. In order reduce implementation cost switch, we aim at minimizing required given delay bound. Conventional Birkhoff-von Neumann matrix decomposition N2 - 2N + 2 configurations schedule, which lead very large The existing...

As the microblogging service (such as Weibo) is becoming popular, spam becomes a serious problem of affecting credibility and readability Online Social Networks. Most existing studies took use set features to identify spam, but without consideration overlap dependency among different features. In this study, we investigate detection by analyzing real dataset collections Weibo propose novel hybrid model spammer detection, called SDHM, which utilizing significant features, i.e. user behavior...

As the microblogging service (such as Weibo) is becoming popular, spam becomes a serious problem of affecting credibility and readability Online Social Networks. Most existing studies took use set features to identify spam, but without consideration overlap dependency among different features. In this study, we investigate detection by analyzing real dataset collections Weibo propose novel hybrid model spammer detection, called SDHM, which utilizing significant features, i.e. user behavior...

We consider power allocation in OFDM based underlay cognitive radio networks with partially known inter-system CSI (Channel State Information). Under a given total transmit limit at the SU (Secondary User) transmitter, goal is to assign certain amount of for signal transmission each sub-channel, such that SU's overall throughput can be maximized, and average interference PU kept within target outage probability level. The existing algorithm adopts an iterative binary searching process find...

We consider traffic scheduling in performance guaranteed switches with optical fabrics to ensure 100% throughput and bounded packet delay. Each switch reconfiguration consumes a constant period of time called <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">reconfiguration</i> xmlns:xlink="http://www.w3.org/1999/xlink">overhead</i> , during which no can be transmitted across the switch. To minimize delay bound for an arbitrary matrix, number...

We consider energy detection based spectrum sensing for opportunistic SU (Secondary User) transmissions in cognitive radio networks. Due to the time-varying nature of wireless fading channels and PU (Primary activities, instantaneous SINR (Signal Interference plus Noise Ratio) at receiver changes from slot a time-slotted system. Unlike conventional detector which uses fixed value threshold detect PU's occurrence, we let transmitter dynamically adjust according SINR. Under constraint limiting...

With the wide deployment of 3G/4G cellular data networks there is a tremendous growth mobile Internet access worldwide. We conduct detailed measurement study about user behaviors, application usage and location patterns users. present methodology that correlates different network attributes information together. For behaviors we classify all users into four groups by time division investigate types example, find midnight consume bandwidth. categorize applications several provide insights...

Network-wide virtualization is a promising technique that will enable future Internet to support variety of network services and architectures over shared optical substrate. In network-wide virtualization, it important efficiently map virtual infrastructure (VI) wide-area guarantee the survivability such VI against failures. this paper, we study regional failure-resilient mapping (RRVIM) problem, where failure-dependent protection approach used efficient sharing resources among different...

Multi-label classification refers to the task of predicting potentially multiple labels for a given instance. Conventional multi-label approaches focus on single objective setting, where learning algorithm optimizes over performance criterion (e.g., Ranking Loss ) or heuristic function. The basic assumption is that optimization one can improve overall and meet requirements various applications. However, in many real applications, an optimal classifier may need consider trade-offs among...

In order to enhance software reliability, defect prediction is used predict potential defects and improve efficiency of examination. Traditional methods mainly focus on design static code metrics, building machine learning classifiers pieces that potentially defective. However, these manual extracted features do not contain syntactic semantic information programs. These much more important than those metrics can the accuracy prediction. this paper, we propose a framework called via...

Femtocell networks are widely being deployed to extend cellular network coverage in indoor environments such as office building spaces and homes. In order mitigate possible co-channel interferences, designs based on the concept of cognitive radio (CR) that enables an overlay between macrocell (primary) femtocells (secondary) has been considered a promising approach. The paper first introduces general dynamic sensing mechanism, which is characterized by performing per-time-slot fast upon...

High-performance scalable data centers call for huge-capacity optical packet switches and QoS (quality-of-service) guaranteed multistage switching, where reconfiguration overheads of fabrics efficiency scheduling algorithms are two key challenges. We consider switching in Clos networks with crossbar fabrics, spatial speedup can be implemented using multiple parallel-working middle-stage to fight against switch overhead. By following the existing traffic matrix decomposition approach, our...

This study proposes an approach for achieving strong communication security based on explosive fountain code with coset pre‐coding, where both main and wire‐tap channels are memoryless binary erasure channels. Coset pre‐coding is used to prevent the eavesdroppers from intercepting confidential information leaked bits. Further, designed ensure reliability low leakage. By this way, proposed can keep when probability of channel slightly lower than that channel. Extensive simulations conducted...