This study proposes a novel machine learning architecture and various learning algorithms to build-in anti-phishing services for avoiding cyber-phishing attack. For the rapid develop of information technology, hackers engage in cyber-phishing attack to steal important personal information, which draws information security concerns. The prevention of phishing website involves in various aspect, for example, user training, public awareness, fraudulent phishing, etc. However, recent phishing research has mainly focused on preventing fraudulent phishing and relied on manual identification that is inefficient for real-time detection systems. In this study, we used methods such as ANOVA, X², and information gain to evaluate features. Then, we filtered out the unrelated features and obtained the top 28 most related features as the features to use for the training and evaluation of traditional machine learning algorithms, such as Support Vector Machine (SVM) with linear or rbf kernels, Logistic Regression (LR), Decision tree, and K-Nearest Neighbor (KNN). This research also evaluated the above algorithms with the ensemble learning concept by combining multiple classifiers, such as Adaboost, bagging, and voting. Finally, the eXtreme Gradient Boosting (XGBoost) model exhibited the best performance of 99.2%, among the algorithms considered in this study.

Intrusion detection system (IDS) is a device or software to monitor a network system for malicious activity. In terms of detection results, there could be two types of false, namely, the false positive (FP) which incorrectly detects normal traffic as abnormal, and the false negative (FN) which incorrectly judges malicious traffic as normal. To protect the network system, we expect that FN should be minimized as low as possible. However, since there is a trade-off between FP and FN when IDS detects malicious traffic, it is difficult to reduce the both metrics simultaneously. In this paper, we propose a sequential classifiers combination method to reduce the effect of the trade-off. The single classifier suffers a high FN rate in general, therefore additional classifiers are sequentially combined in order to detect more positives (reduce more FN). Since each classifier can reduce FN and does not generate much FP in our approach, we can achieve a reduction of FN at the final output. In evaluations, we use NSL-KDD dataset, which is an updated version of KDD Cup'99 dataset. WEKA is utilized as a classification tool in experiment, and the results show that the proposed approach can reduce FN while improving the sensitivity and accuracy.

In this paper, we propose a method for automatic virtual resource allocation by using a multi-target classification-based scheme (MTCAS). In our method, an Infrastructure Provider (InP) bundles its CPU, memory, storage, and bandwidth resources as Network Elements (NEs) and categorizes them into several types in accordance to their function, capabilities, location, energy consumption, price, etc. MTCAS is used by the InP to optimally allocate a set of NEs to a Virtual Network Operator (VNO). Such NEs will be subject to some constraints, such as the avoidance of resource over-allocation and the satisfaction of multiple Quality of Service (QoS) metrics. In order to achieve a comparable or higher prediction accuracy by using less training time than the available ensemble-based multi-target classification (MTC) algorithms, we propose a majority-voting based ensemble algorithm (MVEN) for MTCAS. We numerically evaluate the performance of MTCAS by using the MVEN and available MTC algorithms with synthetic training datasets. The results indicate that the MVEN algorithm requires 70% less training time but achieves the same accuracy as the related ensemble based MTC algorithms. The results also demonstrate that increasing the amount of training data increases the efficacy ofMTCAS, thus reducing CPU and memory allocation by about 33% and 51%, respectively.

ECN, as a decisive approach for TCP congestion control, has been proposed for many years. However, its deployment on the Internet is much slower than expected. In this paper, we investigate the state of the deployment of ECN (Explicit Congestion Notification) on the Internet from a different viewpoint. We use the data set of web domains published by Alexa as the hosts to be tested. We negotiate an ECN-Capable and a Not ECN-Capable connections with each host and collect all packets belonging to the connections. By analyzing the header fields of the TCP/IP packets, we dig out the deployment rate, connectivity, variation of round-trip time and time to live between the Not ECN-Capable and ECN-Capable connections as well as the rate of IPv6-Capable web servers. Especially, it is clear that the connectivity is different from the domains (regions on the Internet). We hope that the findings acquired from this study would incentivize ISPs and administrators to enable ECN in their network systems.

Wireless sensor and actuator networks (WSANs) are expected to become key technologies supporting machine-to-machine (M2M) communication in the Internet of things (IoT) era. However, sensors must be able to provide high demand response (DR) levels despite severely limited battery power. Therefore, as part of efforts to achieve a high DR, we are working on research and development related to radio-on-demand sensor and actuator networks (ROD-SANs). ROD-SAN nodes are equipped with wake-up receivers that allow all nodes to stay in sleep mode for a long period of time, and transmit only after the receiver receives a wake-up signal. In addition, sender nodes can direct the receiver nodes to switch communication channels because the wake-up signal also includes information on the channel to use for communication between each other. However, as the number of nodes utilizing the same channel increases, frequent packet collisions occur, thereby degrading response performance. To reduce packet collisions, we propose an own-channel-utilization based channel switching (OCS) scheme, which is a modification of the average-channel-utilization based switching (ACS) as our previous works. The OCS scheme decides whether or not to switch channels based on a probability value that considers not only average-channel utilization of nearby nodes but also own-channel utilization. This approach permits node switching to other channels by considering the overall utilization states of all channels. In this paper, based on simulations, we show that our scheme can improve the delivery ratio by approximately 15% rather than ACS scheme.

In this paper, we propose a method to construct a scalable sensor data stream delivery system that guarantees the specified delivery quality of service (i.e., total reachability to destinations), even when delivery server resources (nodes) are in a heterogeneous churn situation. A number of P2P-based methods have been proposed for constructing a scalable and efficient sensor data stream system that accommodates different delivery cycles by distributing communication loads of the nodes. However, no existing method can guarantee delivery quality of service when the nodes on the system have a heterogeneous churn rate. As an extension of existing methods, which assign relay nodes based on the distributed hashing of the time-to-deliver, our method specifies the number of replication nodes, based on the churn rate of each node and on the relevant delivery paths. Through simulations, we confirmed that our proposed method can guarantee the required reachability, while avoiding any increase in unnecessary resource assignment costs.

Service function chain (SFC) is a series of ordered virtual network functions (VNFs) for processing traffic flows in the virtualized networking environment of future networks. In this paper, we present a mathematical model and dynamic programing based scheme for solving the problem of SFC placement on substrate networks equipped with network function virtualization (NFV) capability. In this paper, we first formulate the overall cost of SFC placement as the combination of setup cost and operation cost. We then formulate the SFC placement problem as an integer linear programing (ILP) model with the objective of minimizing the overall cost of setup and operation, and propose a delay aware dynamic programming based SFC placement scheme for large networks. We conduct numeric simulations to evaluate the proposed scheme. We analyze the cost and performance of network under different optimization objectives, with and without keeping the order of VNFs in SFC. We measure the success rate, resources utilization, and end to end delay of SFC on different topologies. The results show that the proposed scheme outperforms other related schemes in various scenarios.

Sharing perceptual data (e.g., camera and LiDAR data) with other vehicles enhances the traffic safety of autonomous vehicles because it helps vehicles locate other vehicles and pedestrians in their blind spots. Such safety applications require high throughput and short delay, which cannot be achieved by conventional microwave vehicular communication systems. Therefore, millimeter-wave (mmWave) communications are considered to be a key technology for sharing perceptual data because of their wide bandwidth. One of the challenges of data sharing in mmWave communications is broadcasting because narrow-beam directional antennas are used to obtain high gain. Because many vehicles should share their perceptual data to others within a short time frame in order to enlarge the areas that can be perceived based on shared perceptual data, an efficient scheduling for concurrent transmission that improves spatial reuse is required for perceptual data sharing. This paper proposes a data sharing algorithm that employs a graph-based concurrent transmission scheduling. The proposed algorithm realizes concurrent transmission to improve spatial reuse by designing a rule that is utilized to determine if the two pairs of transmitters and receivers interfere with each other by considering the radio propagation characteristics of narrow-beam antennas. A prioritization method that considers the geographical information in perceptual data is also designed to enlarge perceivable areas in situations where data sharing time is limited and not all data can be shared. Simulation results demonstrate that the proposed algorithm doubles the area of the cooperatively perceivable region compared with a conventional algorithm that does not consider mmWave communications because the proposed algorithm achieves high-throughput transmission by improving spatial reuse. The prioritization also enlarges the perceivable region by a maximum of 20%.

Network function virtualization (NFV) flexibly provides servoces by virtualizing network functions on a general-purpose server, and attracted research interest in recent years. In NFV environment, providing service chaining, which dynamically connects each network function (virtual network function: VNF), is critical issue. However, as it is challenging to select the optimal sequence of VNF services in the service chain in a decentralized manner, the distances between the VNFs tend to increase, leading to longer communication and processing delays. Furthermore, it has never considered that certain VNFs that can be exchange the order of services with one another. To address this problem, in this paper, we propose a distributed search method for ordered VNFs to reduce delays while considering the load on control server, by exploiting an in-network guidance technology, called Breadcrrmubs, for query messages.

Linux container technology and clusters of the containers are expected to make web services consisting of multiple web servers and a load balancer portable, and thus realize easy migration of web services across the different cloud providers and on-premise datacenters. This prevents service to be locked-in a single cloud provider or a single location and enables users to meet their business needs, e.g., preparing for a natural disaster. However existing container management systems lack the generic implementation to route the traffic from the internet into the web service consisting of container clusters. For example, Kubernetes, which is one of the most popular container management systems, is heavily dependent on cloud load balancers. If users use unsupported cloud providers or on-premise datacenters, it is up to users to route the traffic into their cluster while keeping the redundancy and scalability. This means that users could easily be locked-in the major cloud providers including GCP, AWS, and Azure. In this paper, we propose an architecture for a group of containerized load balancers with ECMP redundancy. We containerize Linux ipvs and exabgp, and then implement an experimental system using standard Linux boxes and open source software. We also reveal that our proposed system properly route the traffics with redundancy. Our proposed load balancers are usable even if the infrastructure does not have supported load balancers by Kubernetes and thus free users from lock-ins.

Low Power Wide Area Network (LPWAN) is designed for low-bandwidth, low-power, long-distance, large-scale connected IoT applications and realistic for networking in an emergency or restricted situation, so it has been proposed as an attractive communication technology to handle unexpected situations that occur during and/or after a disaster. However, the traditional LPWAN with its default protocol will reduce the communication efficiency in disaster situation because a large number of users will send and receive emergency information result in communication jams and soaring error rates. In this paper, we proposed a LPWAN based decentralized network structure as an extension of our previous Disaster Information Sharing System (DISS). Our network structure is powered by Named Node Networking (3N) which is based on the Information-Centric Networking (ICN). This network structure optimizes the excessive useless packet forwarding and path optimization problems with node name routing (NNR). To verify our proposal, we conduct a field experiment to evaluate the efficiency of packet path forwarding between 3N+LPWA structure and ICN+LPWA structure. Experimental results confirm that the load of the entire data transmission network is significantly reduced after NNR optimized the transmission path.

Content-centric networking (CCN) promises efficient content delivery services with in-network caching. However, it cannot utilize cached chunks near users if they are not on the shortest path to the server, and it tends to mostly cache highly popular chunks in a domain. This degrades cache efficiency in obtaining various contents in CCN. Therefore, we propose hash-based cache distribution and search schemes to obtain various contents from nearby nodes and evaluate the effectiveness of this approach through simulation.

The GUINNESS (GUI based binarized neural network synthesizer) is an open-source tool flow for a binarized deep neural network toward FPGA implementation based on the GUI including both the training on the GPU and inference on the FPGA. Since all the operation is done on the GUI, the software designer is not necessary to write any scripts to design the neural network structure, training behavior, only specify the values for hyperparameters. After finishing the training, it automatically generates C++ codes to synthesis the bit-stream using the Xilinx SDSoC system design tool flow. Thus, our tool flow is suitable for the software programmers who are not familiar with the FPGA design. In our tool flow, we modify the training algorithms both the training and the inference for a binarized CNN hardware. Since the hardware has a limited number of bit precision, it lacks minimal bias in training. Also, for the inference on the hardware, the conventional batch normalization technique requires additional hardware. Our modifications solve these problems. We implemented the VGG-11 benchmark CNN on the Digilent Inc. Zedboard. Compared with the conventional binarized implementations on an FPGA, the classification accuracy was almost the same, the performance per power efficiency is 5.1 times better, as for the performance per area efficiency, it is 8.0 times better, and as for the performance per memory, it is 8.2 times better. We compare the proposed FPGA design with the CPU and the GPU designs. Compared with the ARM Cortex-A57, it was 1776.3 times faster, it dissipated 3.0 times lower power, and its performance per power efficiency was 5706.3 times better. Also, compared with the Maxwell GPU, it was 11.5 times faster, it dissipated 7.3 times lower power, and its performance per power efficiency was 83.0 times better. The disadvantage of our FPGA based design requires additional time to synthesize the FPGA executable codes. From the experiment, it consumed more three hours, and the total FPGA design took 75 hours. Since the training of the CNN is dominant, it is considerable.

In order to realize intelligent robot system, it is required to process large amount of data input from complex and different kinds of sensors in a short time. FPGA is expected to improve process performance of robots due to better performance per power consumption than high performance CPU, but it has lower development productivity than software. In this paper, we discuss automatic generation of FPGA components for robots. A design tool, developed for easy integration of FPGA into robots, is proposed. The tool named cReComp can automatically convert circuit written in Verilog HDL into a software component compliant to a robot software framework ROS (Robot Operation System), which is the standard in robot development. To evaluate its productivity, we conducted a subject experiment. As a result, we confirmed that the automatic generation is effective to ease the development of FPGA components for robots.

We propose an object detector using a sliding window method for an event-driven camera which outputs a subtracted frame (usually a binary value) when changes are detected in captured images. Since sliding window skips unchanged portions of the output, the number of target object area candidates decreases dramatically, which means that our system operates faster and with lower power consumption than a system using a straightforward sliding window approach. Since the event-driven camera output consists of binary precision frames, an all binarized convolutional neural network (ABCNN) can be available, which means that it allows all convolutional layers to share the same binarized convolutional circuit, thereby reducing the area requirement. We implemented our proposed method on the Xilinx Inc. Zedboard and then evaluated it using the PETS 2009 dataset. The results showed that our system outperformed BCNN system from the viewpoint of detection performance, hardware requirement, and computation time. Also, we showed that FPGA is an ideal method for our system than mobile GPU. From these results, our proposed system is more suitable for the embedded systems based on stationary cameras (such as security cameras).

Since the hardware resource of a single FPGA is limited, one idea to scale the performance of FPGA-based HPC applications is to expand the design space with multiple FPGAs. This paper presents a scalable architecture of a deeply pipelined stream computing platform, where available parallelism and inter-FPGA link characteristics are investigated to achieve a scaled performance. For a practical exploration of this vast design space, a performance model is presented and verified with the evaluation of a tsunami simulation application implemented on Intel Arria 10 FPGAs. Finally, scalability analysis is performed, where speedup is achieved when increasing the computing pipeline over multiple FPGAs while maintaining the problem size of computation. Performance is scaled with multiple FPGAs; however, performance degradation occurs with insufficient available bandwidth and large pipeline overhead brought by inadequate data stream size. Tsunami simulation results show that the highest scaled performance for 8 cascaded Arria 10 FPGAs is achieved with a single pipeline of 5 stream processing elements (SPEs), which obtained a scaled performance of 2.5 TFlops and a parallel efficiency of 98%, indicating the strong scalability of the multi-FPGA stream computing platform.

With the continuous refinement of Deep Neural Networks (DNNs), a series of deep and complex networks such as Residual Networks (ResNets) show impressive prediction accuracy in image classification tasks. Unfortunately, the structural complexity and computational cost of residual networks make hardware implementation difficult. In this paper, we present the quantized and reconstructed deep neural network (QR-DNN) technique, which first inserts batch normalization (BN) layers in the network during training, and later removes them to facilitate efficient hardware implementation. Moreover, an accurate and efficient residual network accelerator (RNA) is presented based on QR-DNN with batch-normalization-free structures and weights represented in a logarithmic number system. RNA employs a systolic array architecture to perform shift-and-accumulate operations instead of multiplication operations. QR-DNN is shown to achieve a 1∼2% improvement in accuracy over existing techniques, and RNA over previous best fixed-point accelerators. An FPGA implementation on a Xilinx Zynq XC7Z045 device achieves 804.03 GOPS, 104.15 FPS and 91.41% top-5 accuracy for the ResNet-50 benchmark, and state-of-the-art results are also reported for AlexNet and VGG.

This paper is concerned with a mixed-integer optimization (MIO) approach to selecting a subset of relevant features from among many candidates. For ordinal classification, a sequential logit model and an ordered logit model are often employed. For feature subset selection in the sequential logit model, Sato et al.[22] recently proposed a mixed-integer linear optimization (MILO) formulation. In their MILO formulation, a univariate nonlinear function contained in the sequential logit model was represented by a tangent-line-based approximation. We extend this MILO formulation toward the ordered logit model, which is more commonly used for ordinal classification than the sequential logit model is. Making use of tangent planes to approximate a bivariate nonlinear function involved in the ordered logit model, we derive an MILO formulation for feature subset selection in the ordered logit model. Our computational results verify that the proposed method is superior to the L1-regularized ordered logit model in terms of solution quality.

Facial micro-expression is momentary and subtle facial reactions, and it is still challenging to automatically recognize facial micro-expression with high accuracy in practical applications. Extracting spatiotemporal features from facial image sequences is essential for facial micro-expression recognition. In this paper, we employed 3D Convolutional Neural Networks (3D-CNNs) for self-learning feature extraction to represent facial micro-expression effectively, since the 3D-CNNs could well extract the spatiotemporal features from facial image sequences. Moreover, transfer learning was utilized to deal with the problem of insufficient samples in the facial micro-expression database. We primarily pre-trained the 3D-CNNs on normal facial expression database Oulu-CASIA by supervised learning, then the pre-trained model was effectively transferred to the target domain, which was the facial micro-expression recognition task. The proposed method was evaluated on two available facial micro-expression datasets, i.e. CASME II and SMIC-HS. We obtained the overall accuracy of 97.6% on CASME II, and 97.4% on SMIC, which were 3.4% and 1.6% higher than the 3D-CNNs model without transfer learning, respectively. And the experimental results demonstrated that our method achieved superior performance compared to state-of-the-art methods.

A new fuzzy level set method (FLSM) based on the global search capability of quantum particle swarm optimization (QPSO) is proposed to improve the stability and precision of image segmentation, and reduce the sensitivity of initialization. The new combination of QPSO-FLSM algorithm iteratively optimizes initial contours using the QPSO method and fuzzy c-means clustering, and then utilizes level set method (LSM) to segment images. The new algorithm exploits the global search capability of QPSO to obtain a stable cluster center and a pre-segmentation contour closer to the region of interest during the iteration. In the implementation of the new method in segmenting liver tumors, brain tissues, and lightning images, the fitness function of the objective function of QPSO-FLSM algorithm is optimized by 10% in comparison to the original FLSM algorithm. The achieved initial contours from the QPSO-FLSM algorithm are also more stable than that from the FLSM. The QPSO-FLSM resulted in improved final image segmentation.

This paper proposes an enhanced affinity graph (EA-graph) for image segmentation. Firstly, the original image is over-segmented to obtain several sets of superpixels with different scales, and the color and texture features of the superpixels are extracted. Then, the similarity relationship between neighborhood superpixels is used to construct the local affinity graph. Meanwhile, the global affinity graph is obtained by sparse reconstruction among all superpixels. The local affinity graph and global affinity graph are superimposed to obtain an enhanced affinity graph for eliminating the influences of noise and isolated regions in the image. Finally, a bipartite graph is introduced to express the affiliation between pixels and superpixels, and segmentation is performed using a spectral clustering algorithm. Experimental results on the Berkeley segmentation database demonstrate that our method achieves significantly better performance compared to state-of-the-art algorithms.

In this paper, we consider a group testing (GT) problem. We derive a lower bound on the probability of error for successful decoding of defected binary signals. To this end, we exploit Fano's inequality theorem in the information theory. We show that the probability of error is bounded as an entropy function, a density of a pooling matrix and a sparsity of a binary signal. We evaluate that for decoding of highly sparse signals, the pooling matrix is required to be dense. Conversely, if dense signals are needed to decode, the sparse pooling matrix should be designed to achieve the small probability of error.

An optimized low-power optical memory access network is proposed to alleviate the cost of microring resonators (MRs) in kilocore systems, such as the pass-by loss and integration difficulty. Compared with traditional electronic bus interconnect, the proposed network reduces power consumption and latency by 80% to 89% and 21% to 24%. Moreover, the new network decreases the number of MRs by 90.6% without an increase in power consumption and latency when making a comparison with Optical Ring Network-on-Chip (ORNoC).

Digital phase-locked loops (DPLLs) have been designed in a number of ways to correctly generate pulse signals in various systems. However, the existing DPLLs have poor acquisition performance or are prone to the divergence phenomenon when modeling and/or round-off errors exist and the noise statistics are incorrect. In this paper, we propose a novel DPLL whose phase estimator is designed in hybrid form that utilizes the advantages of Monte Carlo estimation, which is robust to nonlinear effects such as measurement quantization, and a finite memory estimator, which is robust against incorrect noise information and system model mismatch. The robustness of the proposed hybrid Monte Carlo/finite memory DPLL is demonstrated by comparing its phase estimation performance via a numerical example.

In this paper, we propose a new dynamic load balancing scheme according to load threshold adjustment and incentives mechanism. The proposed scheme adjusts the load threshold of a node by comparing it with a mean threshold of adjacent nodes, thereby increasing the threshold evenly. We also assign the incentives and penalties to each node through a comparison of the mean threshold of all the nodes in order to increase autonomous load balancing participation.

Vehicle detection is challenging in natural traffic scenes because there exist a lot of occlusion. Because of occlusion, detector's training strategy may lead to mismatch between features and labels. As a result, some predicted bounding boxes may shift to surrounding vehicles and lead to lower confidences. These bounding boxes will lead to lower AP value. In this letter, we propose a new approach to address this problem. We calculate the center of visible part of current vehicle based on road information. Then a variable-radius Gaussian weight based method is applied to reweight each anchor box in loss function based on the center of visible part in training time of SSD. The reweighted method has ability to predict higher confidences and more accurate bounding boxes. Besides, the model also has high speed and can be trained end-to-end. Experimental results show that our proposed method outperforms some competitive methods in terms of speed and accuracy.

A novel image enhancement method for vein recognition is introduced. Inspired by observation that the intensity of the vein vessel changes rapidly during the smoothing process compared to that of background (i.e., skin tissue) due to its thin and long shape, we propose to exploit the smoothing speed as a restoration weight for the vein image enhancement. Experimental results based on the CASIA multispectral palm vein database demonstrate that the proposed method is effective to improve the performance of vein recognition.

To expand the use of systems using a camera on portable devices such as tablets and smartphones, we have developed and propose a memory saving feature descriptor, the use of which is one of the essential techniques in computer vision. The proposed descriptor compares pixel values of pre-fixed positions in the small patch around the feature point and stores binary values. Like the conventional descriptors, it extracts the patch on the basis of the scale and orientation of the feature point. For memories of the same size, it achieves higher accuracy than ORB and BRISK in all cases and AKAZE for the images with textured regions.

Saliency quality assessment aims at estimating the objective quality of a saliency map without access to the ground-truth. Existing works typically evaluate saliency quality by utilizing information from saliency maps to assess its compactness and closedness while ignoring the information from image content which can be used to assess the consistence and completeness of foreground. In this letter, we propose a novel multi-information fusion network to capture the information from both the saliency map and image content. The key idea is to introduce a siamese module to collect information from foreground and background, aiming to assess the consistence and completeness of foreground and the difference between foreground and background. Experiments demonstrate that by incorporating image content information, the performance of the proposed method is significantly boosted. Furthermore, we validate our method on two applications: saliency detection and segmentation. Our method is utilized to choose optimal saliency map from a set of candidate saliency maps, and the selected saliency map is feeded into an segmentation algorithm to generate a segmentation map. Experimental results verify the effectiveness of our method.

Real-time monitoring of heart rate (HR) and body temperature (BT) is crucial for the prognosis and the diagnosis of cardiovascular disease and healthcare. Since current monitoring systems are too rigid and bulky, it is not easy to attach them to the human body. Also, their large current consumption limits the working time. In this paper, we develop a wireless sensor patch for HR and BT by integrating sensor chip, wireless communication chip, and electrodes on the flexible boards that is covered with non-toxic, but skin-friendly adhesive patch. Our experimental results reveal that the flexible wireless sensor patch can efficiently detect early diseases by monitoring the HR and BT in real time.