Showing papers in "Chinese Journal of Electronics in 2019"
TL;DR: Compared with other methods, the CCLA model can well capture the local and long distance semantic and emotional information and shows superior performances over several state-of-the-art baseline methods.
Abstract: The major challenge that text sentiment classification modeling faces is how to capture the intrinsic semantic, emotional dependence information and the key part of the emotional expression of text. To solve this problem, we proposed a Coordinated CNN-LSTM-Attention(CCLA) model. We learned the vector representations of sentence with CCLA unit. Semantic and emotional information of sentences and their relations are adaptively encoded to vector representations of document. We used softmax regression classifier to identify the sentiment tendencies in the text. Compared with other methods, the CCLA model can well capture the local and long distance semantic and emotional information. Experimental results demonstrated the effectiveness of CCLA model. It shows superior performances over several state-of-the-art baseline methods.
53 citations
TL;DR: This article focuses on mining useful data from the massive online education data, by using transfer learning and Hadoop, to construct Online education data classification framework (OEDCF), and design an algorithm Tr_MAdaBoost, which overcomes the traditional classification algorithms.
Abstract: With the rapid development of network information technology and the wide application of smart phones, tablet PCs and other mobile terminals, online education plays an increasingly important role in social life. This article focuses on mining useful data from the massive online education data, by using transfer learning, relying on Hadoop, to construct Online education data classification framework (OEDCF), and design an algorithm Tr_MAdaBoost. This algorithm overcomes the traditional classification algorithms in which the required data must be restricted to independent and identically distributed data, since online education using this new algorithm can achieve the correct classification even it has different data distribution. At the same time, with the help of Hadoop's parallel processing architecture, OEDCF can greatly enhance the efficiency of data processing, create favorable conditions for learning analysis, and promote personalized learning and other activities of big data era.
29 citations
TL;DR: In this paper, an improved distance vector hop (DV-Hop) localization algorithm is proposed to ensure accurate node localization in WSNs, where a probability information based selective strategy for the selection of beacon nodes is proposed.
Abstract: Node Localization is a fundamental issue for many critical applications in Wireless sensor networks (WSNs). Traditional DV-Hop localization algorithm and corresponding improved ones still cannot provide su.cient localization accuracy in such WSNs. To ensure accurate localization, this paper proposes an improved DistancevectorHop (DV-Hop) localization algorithm. Under such an algorithm, we determine a corrected average hopdistance of beacon nodes by employing the di.erences between actual and estimated distance among beacon nodes in WSNs. We propose a probability information based selective strategy for the selection of beacon nodes. Based on these selected beacon nodes, we adopt a two dimensional hyperbolic function to predict the locations of unknown nodes. Simulation results are provided to illustrate the localization accuracy of our algorithm compared with traditional DV-Hop algorithm and its two improved algorithms in WSNs.
28 citations
TL;DR: This work proposes a novel mechanism as defense solution for ARP spoofing oriented to OpenFlow platform and shows that this solution can reduce the security threat of ARp spoofing remarkably on Open Flow platform and related SDN platforms.
Abstract: As an emerging network technology, Software-defined network (SDN), has been rapidly developing for recent years due to its advantage in network management and updating. There are still a lot of open problems while applying this novel technology in reality, especially for meeting security demands. The Address resolution protocol (ARP) spoofing, a representative network attack in traditional networks is investigated. We implement the ARP spoofing in SDN network firstly and find that the threat of ARP attack still exists and has big impact on the network. We propose a novel mechanism as defense solution for ARP spoofing oriented to OpenFlow platform. Theoretical analyzation is given, and the mechanism is implemented as a module of POX controller. Experiment results and performance evaluations show that our solution can reduce the security threat of ARP spoofing remarkably on OpenFlow platform and related SDN platforms.
21 citations
TL;DR: This paper analyzed the existing network security situation evaluation methods and discovered that they cannot accurately reflect the features of large-scale, synergetic, multi-stage gradually shown by network attack behaviors.
Abstract: This paper analyzed the existing network security situation evaluation methods and discovered that they cannot accurately reflect the features of large-scale, synergetic, multi-stage gradually shown by network attack behaviors. For this purpose, the association between attack intention and network configuration information was deep analyzed. Then a network security situation evaluation method based on attack intention recognition was proposed. Unlike traditional method, the evaluation method was based on intruder. This method firstly made causal analysis of attack event and discovered and simplified intrusion path to recognize every attack phases, then realized situation evaluation based on the attack phases. Lastly attack intention was recognized and next attack phase was forecasted based on achieved attack phases, combined with vulnerability and network connectivity. A simulation experiments for the proposed network security situation evaluation model is performed by network examples. The experimental results show that this method is more accurate on reflecting the truth of attack. And the method does not need training on the historical sequence, so the method is more effective on situation forecasting.
20 citations
TL;DR: Proven and analysis show that the proposed self-certified cross-cluster Asymmetric group key agreement (SC-AGKA) has the advantages of in security and energy consumption.
Abstract: Wireless sensor networks have some obvious characteristics, such as communication range is limited, computing power is limited and energy is limited Group key agreement in this environment requires a cross-cluster, computation and communication overhead are lightweight and highly safe group key agreement protocol Aiming at these demands, the paper proposes a Self-certified cross-cluster Asymmetric group key agreement (SC-AGKA) To establish a lightweight and efficient group communication channel among sensor nodes According to the cluster head as the bridge node to realize the sensor nodes in different cluster have the same group key information, and negotiate a pair of asymmetric group keys to realize the cross cluster secure communication The group communication adopts asymmetric encryption mechanism It realizes the group security communication mechanism of message sender unconstraint The asymmetric group key agreement has the key self-certified, which does not need additional rounds to verify the correctness of group key Proven and analysis show that the proposed protocol has the advantages of in security and energy consumption
18 citations
TL;DR: The key insight of DBNPM is Deep belief network (DBN) technology, which is an effective deep learning technique in image processing and natural language processing, whose features are similar to defects in source program.
Abstract: Defect distribution prediction is a meaningful topic because software defects are the fundamental cause of many attacks and data loss. Building accurate prediction models can help developers find bugs and prioritize their testing efforts. Previous researches focus on exploring different machine learning algorithms based on the features that encode the characteristics of programs. The problem of data redundancy exists in software defect data set, which has great influence on prediction effect. We propose a defect distribution prediction model (Deep belief network prediction model, DBNPM), a system for detecting whether a program module contains defects. The key insight of DBNPM is Deep belief network (DBN) technology, which is an effective deep learning technique in image processing and natural language processing, whose features are similar to defects in source program. Experiment results show that DBNPM can efficiently extract and process the data characteristics of source program and the performance is better than Support vector machine (SVM), Locally linear embedding SVM (LLE-SVM), and Neighborhood preserving embedding SVM (NPE-SVM).
13 citations
TL;DR: A novel method is proposed based on deep learning for classifying the five types of incredible messages on social media based on three dimensions of information evaluation metrics and a series of experiments demonstrate that the proposed method outperforms the state-of-the-art methods.
Abstract: How to classify incredible messages has attracted great attention from academic and industry nowadays. The recent work mainly focuses on one type of incredible messages (a.k.a rumors or fake news) and achieves some success to detect them. The existing problem is that incredible messages have different types on social media, and rumors or fake news cannot represent all incredible messages. Based on this, in the paper, we divide messages on social media into five types based on three dimensions of information evaluation metrics. And a novel method is proposed based on deep learning for classifying the five types of incredible messages on social media. More specifically, we use attention mechanism to obtain deep text semantic features and strengthen emotional semantics features, meanwhile, construct universal metadata as auxiliary features, concatenating them for incredible messages classification. A series of experiments on two representative real-world datasets demonstrate that the proposed method outperforms the state-of-the-art methods.
13 citations
TL;DR: A novel community detection algorithm called the Deep auto-encoded clustering algorithm (DAC), in which unsupervised and sparse single autoencoders are trained and piled up one after another to embed key community information in a lower-dimensional representation, such that it can be handled easier by clustering strategies.
Abstract: The prevalence of deep learning has inspired innovations in numerous research fields including community detection, a cornerstone in the advancement of complex networks. We propose a novel community detection algorithm called the Deep auto-encoded clustering algorithm (DAC), in which unsupervised and sparse single autoencoders are trained and piled up one after another to embed key community information in a lower-dimensional representation, such that it can be handled easier by clustering strategies. Extensive comparison tests undertaken on synthetic and real world networks reveal two advantages of the proposed algorithm: on the one hand, DAC shows higher precision than the k
-means community detection method benefiting from the integration of sparsity constraints. On the other hand, DAC runs much faster than the spectral community detection algorithm based on the circumvention of the time-consuming eigenvalue decomposition procedure.
12 citations
TL;DR: This paper tries to list several representative issues of this research topic, and briefly describe their recent research progress and some related works proposed along this research line.
Abstract: Deep learning has been attracting increasing attention in the recent decade throughout science and engineering due to its wide range of successful applications. In real problems, however, most implementation stages for applying deep learning still require inevitable manual interventions, which naturally conducts difficulty in its availability to general users with less expertise and also deviates from the intelligence of humans. It is thus a challenging while critical issue to enhance the level of automation across all elements of the entire deep learning framework, like input amelioration, model designing and learning, and output adjustment. This paper tries to list several representative issues of this research topic, and briefly describe their recent research progress and some related works proposed along this research line. Some specific challenging problems have also been presented.
12 citations
TL;DR: This work proposes a public auditing scheme for data confidentiality, in which user resorts to a Third-party auditor (TPA) for auditing, and design a special log called attestation in which hash user pseudonym is used to preserve user privacy.
Abstract: Cloud data confidentiality need to be audited for the data owner's concern. Confidentiality auditing is usually based on logging schemes, whereas cloud data dynamics and sharing group dynamics result in massive logs, which makes confidentiality auditing a formidable task for user with limited resources. So we propose a public auditing scheme for data confidentiality, in which user resorts to a Third-party auditor (TPA) for auditing. Our scheme design a special log called attestation in which hash user pseudonym is used to preserve user privacy. Attestation-based data access identifying is presented in our scheme which brings no new vulnerabilities toward data confidentiality and no extra online burden for user. We further support accountability of responsible user for data leakage based on user pseudonym. Extensive security and performance analysis compare our scheme with existing auditing schemes. Results indicate that the proposed scheme is provably secure and highly efficient.
TL;DR: The results demonstrate that the FTZNN model is a more effective solution model for solving time-varying quadratic minimization problems.
Abstract: This paper proposes a Finite-time Zhang neural network (FTZNN) to solve time-varying quadratic minimization problems. Different from the original Zhang neural network (ZNN) that is specially designed to solve time-varying problems and possesses an exponential convergence property, the proposed neural network exploits a sign-bi-power activation function so that it can achieve the finite-time convergence. In addition, the upper bound of the finite convergence time for the FTZNN model is analytically estimated in theory. For comparative purposes, the original ZNN model is also presented to solve time-varying quadratic minimization problems. Numerical experiments are performed to evaluate and compare the performance of the original ZNN model and the FTZNN model. The results demonstrate that the FTZNN model is a more effective solution model for solving time-varying quadratic minimization problems.
TL;DR: The improved algorithm has better performance of optimization, faster convergence speed and higher convergence accuracy, and theoretical analysis proved the convergence and time complexity of the improved algorithm.
Abstract: For the shortcomings of the basic flower pollination algorithm, this paper proposes a differential evolution flower pollination algorithm with dynamic switch probability based on the Weibull distribution. This new algorithm improved the convergence rate and precision. The switch probability is improved by Weibull distribution function combined with the number of iterations. It can balance the relationship between the global pollination and the local pollination to improve the overall optimization performance of the algorithm. Random mutation operator is merged into the global pollination process to increase diversity of the population, enhance the ability of the algorithm's global search and avoid premature convergence. In the process of local pollination, directed mutation and crossover operation of the differential evolution are incorporated, it makes the individual flower position update with the memory function, which can choose the direction of variation reasonably. The use of cross-operation can avoid new solutions crossing the boundary. Convergence rate is improved and the algorithm can approach the global optimal solution continuously. Theoretical analysis proved the convergence and time complexity of the improved algorithm. The simulation results based on the function optimization problem show that the improved algorithm has better performance of optimization, faster convergence speed and higher convergence accuracy.
TL;DR: A novel energy-efficient cooperative strategy in a Radio frequency (RF) energy harvesting cognitive radio network, where the Secondary user provides energy supply and relaying assist for the Primary user (PU) in exchange for the transmission opportunities, is proposed.
Abstract: Energy-efficient cooperative spectrum sharing has become a new trend based on financial and environmental considerations. We focus on a novel energy-efficient cooperative strategy in a Radio frequency (RF) energy harvesting cognitive radio network, where the Secondary user (SU) provides energy supply and relaying assist for the Primary user (PU) in exchange for the transmission opportunities. On the basis of energy and information cooperation, an Energy efficiency (EE) maximization non-convex problem for SU is formulated under the PU's Quality of service (QoS) requirement and energy-causality constraint. With the aid of Dinkebalch's method and convex optimization technique, a joint time and power allocation scheme is proposed. Numerical results show that the proposed scheme achieves higher SU's EE and better PU's QoS guarantee compared to the scheme in which the SU only provides energy for the PU's transmission.
TL;DR: Simulation results demonstrate that the proposed Load-balancing and QoS based dynamic resource allocation method can provide heterogeneous smart grid businesses with differentiated service, improve the utilization and economic benefits of the network and make the network more balanced.
Abstract: Real-time network resource allocation based on virtualization technology is an important method to solve the solidification problem of Fiber-wireless (FiWi) access networks. To increase the resource utilization and meet the specific QoS requirements of smart gird communication services, a Load-balancing and QoS based dynamic resource allocation method (LbQ-DR) is proposed with three sub-mechanisms. A time-window based substrate network resource update mechanism is designed to describe the realtime resource consumption of substrate networks, which can balance the accuracy of resource status and the complexity of the allocation algorithm. A QoS-based Virtual network request (VNR) sorting mechanism is presented to precisely calculate the priority of VNRs and reasonably sort the incoming services. A load-balancing based resource allocation mechanism is designed to avoid unbalanced resource consumption. Especially, the channel interference is considered in the cost of embedding and a collision domain mechanism is introduced to decrease interference. Simulation results demonstrate that the proposed method can provide heterogeneous smart grid businesses with differentiated service, improve the utilization and economic benefits of the network and make the network more balanced.
TL;DR: Experimental results on two public databases for OD segmentation show that the proposed method achieves the state-of-the-art performance.
Abstract: Accurate segmentation of Optic disc (OD) is significant for the automation of retinal analysis and retinal diseases screening. This paper proposes a novel optic disc segmentation method based on the saliency. It includes two stages: optic disc location and saliency-based segmentation. In the location stage, the OD is detected using a matched template and the density of the vessels. In the segmentation stage, we treat the OD as the salient object and formulate it as a saliency detection problem. To measure the saliency of a region, the boundary prior and the connectivity prior are exploited. Geodesic distance to the window boundary is computed to measure the cost the region spends to reach the window boundary. After a threshold and ellipse fitting, we obtain the OD. Experimental results on two public databases for OD segmentation show that the proposed method achieves thestate-of-the-art performance.
TL;DR: The Threat-based typed security π-calculus (πTBTS) is proposed to model declassification and endorsement in mobile computing and solves the problem of the ad-hoc and nondeterministic semantics and builds a bridge between threat assessments and declassification/endorsement.
Abstract: Declassification and endorsement can efficiently improve the usability of mobile applications. However, both declassify and endorse operations in practice are often ad-hoc and nondeterministic, thus, being insecure. From a new perspective of threat assessments, we propose the Threat-based typed security π-calculus (πTBTS) to model declassification and endorsement in mobile computing. Intuitively, when relaxing confidentiality policies and/or integrity policies, we respectively assess threats brought by performing these two relaxes. If these threats are acceptable, the declassification and/or endorsement operations are permitted; Otherwise, they are denied. The proposed assessments have explicit security conditions, results and less open parameters, so our approach solves the problem of the ad-hoc and nondeterministic semantics and builds a bridge between threat assessments and declassification/endorsement.
TL;DR: A novel algorithm to extract common features for each type of emotional states which can reliably present human emotions is proposed, which shows cloud model is potentially useful in pattern recognition and machines learning.
Abstract: Emotions often facilitate interactions among human beings, but the big variation of human emotional states make a negative effect on the reliable emotion recognition. We propose a novel algorithm to extract common features for each type of emotional states which can reliably present human emotions. To uncover the common features from uncertain emotional states, the backward cloud generator is used to discover {Ex, En, He} by integrating randomness and fuzziness. Finally, the proposed method for emotion recognition is verified on the common facial expression datasets, the Extended Cohn-Kanade (CK+) dataset and the Japanese female facial expression (JAFFE). The results are satisfactory, which shows cloud model is potentially useful in pattern recognition and machines learning.
TL;DR: A VAE-GAN based hashing framework for fast image retrieval that combines a Variational autoencoder (VAE) with a Generative adversarial network (GAN) to generate content preserving images for pairwise hashing learning.
Abstract: Inspired by the recent advances in generative networks, we propose a VAE-GAN based hashing framework for fast image retrieval. The method combines a Variational autoencoder (VAE) with a Generative adversarial network (GAN) to generate content preserving images for pairwise hashing learning. By accepting real image and systhesized image in a pairwise form, a semantic perserving feature mapping model is learned under a adversarial generative process. Each image feature vector in the pairwise is converted to a hash codes, which are used in a pairwise ranking loss that aims to preserve relative similarities on images. Extensive experiments on several benchmark datasets demonstrate that the proposed method shows substantial improvement over the state-of-the-art hashing methods.
TL;DR: When n is an odd prime, using random coding, the obtained families of asymptotically good self-dual and LCD codes of length 6n over 𝔽 q are obtained.
Abstract: In this paper, we study double circulant codes of length 2n over the non-chain ring R = 𝔽 q + v𝔽 q + v 2𝔽 q ; where q is an odd prime power and v 3 = v. Exact enumerations of self-dual and LCD double circulant codes of length 2n over R are derived. When n is an odd prime, using random coding, we obtain families of asymptotically good self-dual and LCD codes of length 6n over 𝔽 q .
TL;DR: A novel and scalable detection algorithm based on deep neural networks-an improved Faster Regionbased Convolutional neural networks (Faster R-CNN) by increasing the fusion of feature maps at different levels is proposed, which can be employed to detect and locate polyps, and even achieve a multi-object task for polyps in the future.
Abstract: The deficiencies of existing polyp detection methods remain: i) They primarily depend on the manually extracted features and require considerable amounts of preprocessing. ii) Most traditional methods cannot specify the location of the polyps in colonoscopy images, especially for the polyps with variable size. In order to derive the improvement and lift the accuracy, we propose a novel and scalable detection algorithm based on deep neural networks-an improved Faster Regionbased Convolutional neural networks (Faster R-CNN)-by increasing the fusion of feature maps at different levels. It can be employed to detect and locate polyps, and even achieve a multi-object task for polyps in the future. The experimental consequences demonstrate that the best version among improved algorithms achieves 97.13% accuracy on the CVC-ClinicDB database, overtaking the previous methods.
TL;DR: This paper designs a convolutional neural network model with low computational cost and high classification accuracy, inspired by the multi-visual mechanism of the organism and DenseNet, and adds the attention mechanism of SE-Net.
Abstract: To solve the red jujube classification problem, this paper designs a convolutional neural network model with low computational cost and high classification accuracy. The architecture of the model is inspired by the multi-visual mechanism of the organism and DenseNet. To further improve our model, we add the attention mechanism of SE-Net. We also construct a dataset which contains 23,735 red jujube images captured by a jujube grading system. According to the appearance of the jujube and the characteristics of the grading system, the dataset is divided into four classes: invalid, rotten, wizened and normal. The numerical experiments show that the classification accuracy of our model reaches to 91.89%, which is comparable to DenseNet-121, InceptionV3, InceptionV4, and Inception-ResNet v2. Our model has real-time performance.
TL;DR: New compression entropy criteria for identifying chaotic signal complexity in periodic, quasi-periodic or chaotic state, in mapping results in 3s-graph with significant different shape of good or bad spring and in Construction creep rate with distinguishable value-range.
Abstract: Signal complexity denotes the intricate patterns hidden in the complicated dynamics merging from nonlinear system concerned. The chaotic signal complexity measuring in principle combines both the information entropy of the data under test and the geometry feature embedded. Starting from the information source of Shannon's entropy, combined with understanding the merits and demerits of 0-1 test for chaos, we propose new compression entropy criteria for identifying chaotic signal complexity in periodic, quasi-periodic or chaotic state, in mapping results in 3s-graph with significant different shape of good or bad spring and in Construction creep (CC) rate with distinguishable value-range of [0, 7%], (7%, 50%] or (50%, 84%]. The employed simulation cases are Lorenz, Li and He equations' evolutions, under key information extracting rules of both two-layer compression functions and self-similarity calcu-lation, compared with methods of 0-1 test for chaos, Lyapunov exponent and Spectral Entropy complexity. The research value of this work will provide deep thinking of the concise featureexpressions of chaotic signal complexity measure in feature domain.
TL;DR: The experimental results show the proposed Quaternion locality preserving projection (QLPP) achieves much better performance than the unimodal biometric algorithms, the traditional feature level fusion methods and two quaternion representation methods.
Abstract: This paper proposed Quaternion locality preserving projection (QLPP) for multi-feature multimodal biometric recognition. Multi-features fill the real part or the three imaginary parts of quaternion to constitute the quaternion fusion features. In quaternion division ring, QLPP extracts the local information and finds essential manifold structure of the quaternion fusion features. Deferent from Quaternion principal component analysis (QPCA) and Quaternion fisher discriminant analysis (QFDA), QLPP takes advantage of the optimal linear approximations to find the nonlinear manifold structures. Two experiments are designed: one fuses four features from two biometric modalities, and the other fuses three features from three biometric modalities. The experimental results show the proposed algorithm achieves much better performance than the unimodal biometric algorithms, the traditional feature level fusion methods(weighted sum rule and series rule) and two quaternion representation methods(QPCA and QFDA).
TL;DR: This work proposes a global/local attention method that can recognize a multi-label image from coarse to fine by mimicking how human-beings observe images, and shows that the method outperforms state-of-the-art methods.
Abstract: Great efforts have been made by using deep neural networks to recognize multi-label images. Since multi-label image classification is very complicated, many studies seek to use the attention mechanism as a kind of guidance. Conventional attention-based methods always analyzed images directly and aggressively, which is difficult to well understand complicated scenes. We propose a global/local attention method that can recognize a multi-label image from coarse to fine by mimicking how human-beings observe images. Our global/local attention method first concentrates on the whole image, and then focuses on its local specific objects. We also propose a joint max-margin objective function, which enforces that the minimum score of positive labels should be larger than the maximum score of negative labels horizontally and vertically. This function further improve our multi-label image classification method. We evaluate the effectiveness of our method on two popular multi-label image datasets (i.e., Pascal VOC and MS-COCO). Our experimental results show that our method outperforms state-of-the-art methods.
TL;DR: By computing the Lorenz equation evolution under the contrast tests of the Poincare section and Lyapunov index, a new chaoscriteria design in symbolic dynamics and data compression principles is visualize and may lay the foundation for further expressing the chaotic appearance of novel signals deep into future brainets.
Abstract: The complexity measures of chaotic or periodic signals are perpetual topics of interest to data scientists. This work adheres to the framework of the traditional 0-1 test for chaos and replaces sine and cosine functions by modified sign functions. The compressive mapping rules chosen are one-threshold of three-value or three-threshold of five-value. In new criteria for chaos in forms of the 3s plot and Ks metric compared with 0-1 test results, the periodic state of data features a short beeline instead of a big ring in the pq plot and signs the nearest zero mark, while the chaotic state signs a simple curve instead of a random-walking shape in the pq plot, and shows the nearest one mark. By computing the Lorenz equation evolution under the contrast tests of the Poincare section and Lyapunov index, we visualize a new chaoscriteria design in symbolic dynamics and data compression principles, and our work may lay the foundation for further expressing the chaotic appearance of novel signals deep into future brainets.
TL;DR: This work uses the highway connections between memory cells in adjacent layers to train a small-footprint highway LSTM-RNNs (HLSTM, short-term memory RNNs), which are deeper and thinner compared to conventional L STM-MNNs and show greater reduction in WER.
Abstract: Long short-term memory RNNs (LSTMRNNs) have shown great success in the Automatic speech recognition (ASR) field and have become the state-ofthe- art acoustic model for time-sequence modeling tasks. However, it is still difficult to train deep LSTM-RNNs while keeping the parameter number small. We use the highway connections between memory cells in adjacent layers to train a small-footprint highway LSTM-RNNs (HLSTM-RNNs), which are deeper and thinner compared to conventional LSTM-RNNs. The experiments on the Switchboard (SWBD) indicate that we can train thinner and deeper HLSTM-RNNs with a smaller parameter number than the conventional 3-layer LSTM-RNNs and a lower Word error rate (WER) than the conventional one. Compared with the counterparts of small-footprint LSTMRNNs, the small-footprint HLSTM-RNNs show greater reduction in WER.
TL;DR: A staggered grid scheme is proposed to reduce both the total memory requirement and the CPU time of generating the corrected near matrix in the FFTbased methods.
Abstract: A staggered grid scheme is proposed to reduce both the total memory requirement and the CPU time of generating the corrected near matrix in the FFTbased methods. Two sets of Cartesian grids are used to project the source points and the field points, respectively. The proposed method does not lower the efficiency of computing far matrix-vector products, compared with the traditional uniform Cartesian grid scheme. Some numerical experiments are provided to demonstrate both the correctness and the efficiency of the proposed method.
TL;DR: In this paper, four core smart driving algorithms are determined by studying the architecture of smart driving algorithm and the security issues of these algorithms are investigated by closely examining the work carried out by the algorithms.
Abstract: With the rapid development of the smart driving technology, the security of smart driving algorithms is becoming more and more important. Four core smart driving algorithms are determined by studying the architecture of smart driving algorithm. These algorithms comprise local path planning, pedestrian detection, lane detection and obstacle detection. The security issues of these algorithms are investigated by closely examining the work carried out by the algorithms. We found that there are vulnerabilities in all four algorithms. These vulnerabilities can cause abnormality and even road accidents for the smart cars. The final experiment shows that the vulnerabilities of these algorithms do exist under certain circumstances and therefore have high security risks. This study will lay a foundation to improve the security of the smart driving system
TL;DR: The evaluation results validate that compared with the polling based controller scheduling method, the proposed one can significantly reduce the connection failure ratio and delay.
Abstract: Distributed Denial of Service (DDoS) attack is a difficult issue which needs to be addressed in Software defined networking (SDN). In order to help the controller to weather out the DDoS attack, an efficient controller scheduling method is proposed. The proposed controller scheduling method uses the normalized waiting time, length and extent of the switch being attacked to choose the request that needs to be processed by the controller. The evaluation results validate that compared with the polling based controller scheduling method, the proposed one can significantly reduce the connection failure ratio and delay.