Abstract:Thanks to abundant spectral resource, millimeter wave (mmWave) communications have become an important research area. Though mmWave communications have achieved great progresses in WLAN, it faces huge challenges in realizing wide ultra-high throughput area coverage. The paper provides a comprehensive survey on mmWave channel measurement and modeling, multiple input and multiple output (MIMO) technology and network framework, and proposes a new densely distributed mmWave massive MIMO system by incorporating new technologies such as large-scale cooperative transmission, cloud computing and distributed storage, which is expected to realize ultra-high throughput wide coverage and support middle-high mobility in mmWave communications.

Abstract:By using the antenna index to transmit information invisibly, spatial modulation (SM) technique can obtain high data transmission rate, and overcome the inter channel interference and synchronization problems effectively, as well as reduce the implementation complexity. Multiple input multiple output (MIMO) technique can significantly improve the capacity and spectrum ef ficiency of wireless communication system. To satisfy the high quality and data rate requirements of communication system, SM-MIMO technique emerges by combining SM and MIMO techniques and thus becomes the main research hotspot in the field of wireless communication. The paper summarizes the current research advancement of SM-MIMO from the basic principles and performance analysis. The prospect of SM-MIMO in application is also addressed. Finally, both the current achievement and the further research prospect in this field are discussed and summarized.

Abstract:The large underutilized spectrum of Millimeter wave (mmWave) can efficiently alleviate the spectrum shortage at lower frequency bands, and the physical size of large-scale antennas systems can be greatly reduced due to the shorter wavelength at mmWave frequencies, which enables mmWave communication to be one of the potentially crucial techniques for 5 G wireless communication systems. Considering the severe path loss of mmWave frequencies, beamforming needs to be utilized in an mmWave system for improving the transmission quality. Due to the high power consumption and high cost of fully digital beamforming methods in mmWave massive MIMO systems, hybrid digital and analog beamforming become an important alternative scheme. Here, we first introduce the current research state of hybrid beamforming schemes for mmWave MIMO systems. Then the system model is built. Finally, the crucial technologies of hybrid beamforming,i.e., channel estimation, codebook design and low-complexity design are described.

Abstract:Polar codes are the first coding schemes that probably achieve the Shannon capacity of memory-less symmetric channels with an explicit construction based on the channel polarization. Moreover, polar codes are adopted for control channels in the fifth generation mobile communication systems (5G). Here, the basic principle of polar codes is briefly introduced, including polar encoding and decoding. Furthermore, a concatenated coding scheme of polar codes and parity-check codes, called the parity-check-concatenated (PCC) polar codes is proposed. The information bits are encoded by an outer parity-check encoder and an inner polar encoder. At the receiver, the parity-check-aided successive cancellation list (PC-aided SCL) algorithm is applied for decoding. Simulation results show that PCC polar codes could have evident performance gains over the cyclic redundancy check-concatenated polar codes without increasing the complexity of encoding and decoding. Therefore, PCC polar codes could meet the requirements of 5G for the error correction performance.

Abstract:With the rapid increasing resolution of display devices in computer, television and mobile phone, there is a consensus in the industry to solve the problem of lacking of display link bandwidth using display stream compression(DSC) technologies. Hence, several display stream compression methods continually appear in recent years, for example JPEG-XS, intra-only coding in H.264/AVC and Dirac (VC-2), HEVC-SCC screen contents coding and DSC of VESA, etc. Among these, the VESA DSC is a widely acceptable standard that is used in display link for a low cost, low delay, and visually lossless lightweight codec. A number of high efficient coding technologies are used in DSC including advanced prediction, indexed color history, plain entropy coding and perfect rate control. The paper overviews these new features and key technologies in DSC standard.

Abstract:With the development of network, network security becomes the key course of computer research. Hacker attacks become more and more frequent. The traditional security products have loopholes. Intrusion detection, as an important means of information security, makes up for the shortcomings of the firewall, provides an effective network intrusion detection measures and protects the network security. However, there are a lot of problems in traditional network intrusion detection. Methods based on machine can detect network intrusion automatically, improve the efficiency of intrusion detection, and reduce the false negative rate and false alarm rate. Here, we first introduce some machine learning algorithms briefly, and then analyze the application of machine learning algorithm in network intrusion detection. Moreover ,we compare the advantages and disadvantages of each algorithm applied in intrusion detection. Finally we summarize the application prospect of machine learning to lay the foundation for the network intrusion detection and prevention system with good performance.

Abstract:For the deficiency of the existing words bag in object recognition. We improve the feature extraction and image representation etc to enhance the accuracy. Firstly, a fixed step size is used and scale-intensive is fixed to extract key points, and then the scale-invariant feature transform (SIFT) and local binry pattern(LBP) around the key points in the grids are extracted to describe the shape features and texture features. K-Means clustering algorithm is introduced to generate a visual dictionary and the local descriptors are encoded by approximated locality constrained linear coding, and max pooling and a histograms are generated using spatial pyramid matching. Both the spatial pyramid histograms are connected, therefore, the feature fusion in the image level is implemented under the words bag. Finally the fusion result is sent to the SVM for classification. Experimental result in public datasets shows that the proposed method can achieve higher recognition accuracy.

Abstract:Aiming that the per-user unitary and rate control(PU2RC) algorithm cannot work well in the low number of users and high signal to noise ratio(SNR) in multi-user multiple-input multiple-output （MU-MIMO）system, a downlink adaptive transmission scheme with reconstructing channel information and searching optimal number of service users at the base station is proposed. The base station firstly reconstructs the downlink channel vector using the norm of channel vector and channel direction information (CDI) fed back by users. Then, the user set corresponding to each codebook matrix is expanded by sharing the users among the similar codebook vectors. Finally, the optimal set of service users is selected to maximize the system sum rate. The proposed scheme improves the probability of multi-user transmission with the low number of users and obtains the tradeoff between the inter-user interference and the number of service users. The sum rate of the proposed system is higher than that of the traditional PU2RC algorithm and the existing improved schemes.

Abstract:The traditional principal curves algorithm can obtain good results on small datasets. But the computing and storage resources of a single node cannot meet the requirements of the extraction of principal curves on massive datasets. Distributed parallel computing is one of the most effective way to solve the problems. Therefore, we proposed a distributed soft K-segments algorithm for principal curves based on MapReduce, named DisSKPC. First, we recursively granulated all the numerical data into information granules to limit each granular size and ensure the relevance of the data in the granules using the distributed K-Means algorithm. Then we calculated the local principal component segments of each granule and eliminated over-fitting segments that may arise in the area of high-density and high-curvature using the noise variance. Finally, we connected these local principal component segments using the Hamiltonian path and greedy algorithm, forming a best curve through the middle of the data cloud. Experimental results demonstrate the feasibility and scalability of the proposed DisSKPC algorithm.

Abstract:Traditional text classification methods assume that feature words in the training set and test set follow the same probability distribution. Nevertheless, deviations exist in a practical application, which can affect the final classification results. To solve the problem, a feature transfer learning algorithm for text categorization is proposed. By calculating the transfer volume and amending the vector space model in the training set, the distribution probability of feature words can be reconciled for the training set and test set. Experiments on Chinese spam filtering and web page classification data sets demonstrate that the proposed method can eliminate the dissimilarity of distributions of feature words, and improve the various indexes of test classification evidently. 

Abstract:Timeline summarization is the process of creating summaries towards topic information and development over time in natural language processing. Some algorithms are proposed to generate summaries towards long text like news, but seldom focus on timeline summaries of short text like microblog. Here, we propose a microblog timeline summarization based on sliding window (MTSW), which simultaneously incorporates content coverage, temporal distribution and influence to evaluate candidate timeline summaries. In the algorithm, representative terms are selected to represent microblog feature according to intensity of terms and entropy. We build a comprehensive indicator for evaluating the timeline summary based on the above three indicators. Then, we use sliding window to generate microblog timeline summary. Experiments on the real-world event datasets verify the effectiveness of the proposed method.

Abstract:To address the problem of sensor-based gait cycle detection method, which needs high cooperation of users. Vision-based method is used to develop an accurate gait cycle detection algorithm. Inspired by the idea of frame difference, a novel gait representation feature, namely frame difference temporal and spatial (FDTS) feature is designed. FDTS contains the temporal and spatial information of gait. Thus it can accurately present all states of gait cycle. A new toe-off detection algorithm is first proposed based on FDTS. Then a gait cycle detection method is presented based on the toe-off detection algorithm. Experiments on the public dataset demonstrate the state-of-the-art performance of the method.

Abstract:In sound source direction of arrival (DOA) estimation, one of the typical methods is based on the time difference of arrival (TDOA). For the TDOA-based sound source DOA estimation, the TDOA-DOA mapping is a crucial step. Here, we propose a TDOA-DOA mapping approach based on the multi-kernel least-squares support vector regression (LS-SVR), and also analyze its performance with sparsification. In addition, we present an outlier detection method based on the normalized median filtering to post-process the TDOA estimation for improving the performance of TDOA-DOA mapping in noisy reverberant environments. Simulation results show that the proposed method is superior to its counterparts, such as LS and single-kernel LS-SVR methods.

Abstract:As a fundamental problem of text categorization, text representation is widely concerned. Currently, there are three main ways of text representation: bag-of-words model, latent semantic representation and knowledge-based explicit semantic representation. The paper analyzes and compared the effects of these methods applied to text categorization. Experiments show that the knowledge-based explicit semantic representation cannot improve the text categorization performance as expected. To tackle the problem that the knowledge-based explicit semantic representation easily introduces noise in extending text, a supervised explicit semantic representation method is proposed. The dataset label information is used to identify the most relevant concepts in document and the document is represented in explicit semantic based on expanding those key concepts. The results of three datasets confirm the effectiveness of the proposed method.

Abstract:The existing global manifold learning algorithms are relatively sensitive to the neighborhood size, which is difficult to select efficiently. The reason is mainly because the neighborhood graph is constructed based on Euclidean distance, by which shortcut edges tend to be introduced into the neighborhood graph. To overcome this problem, a global manifold learning algorithm is proposed based on random walk, called the random walk-based isometric mapping (RW-ISOMAP). Compared with Euclidean distance, the commute time distance, achieved by the random walk on the neighborhood graph, can measure the similarity between the given data within the nonlinear geometric structure to a certain extent, thus it can provide robust results and is more suitable to construct the neighborhood graph. Consequently, by constructing the neighborhood graph based on the commute time distance, RW-ISOMAP is less sensitive to the neighborhood size and more robust than the existing global manifold learning algorithms. Finally, the experiment verifies the effectiveness of RW-ISOMAP.

Abstract:A new algorithm for edge detection is proposed to anti-noisely detect weak edges and broken edges produced in traditional edge detectors. First, edge is extracted and then connected. Edge is the reaction of gray mutation. The gradient of the pixels is composed of weighted pixel number which signifys the amount of pixels adjoin this pixel in the symmetry location. The direction of the pixels with larger gradient is calculated. Edges are extracted based on the continuity of pixels′ direction. To compensate the faults of edge image, we use ant colony algorithm to connect edges. The experimental result shows that the method has strong anti-noise ability, especially for salt and pepper noise, and can effectively detect the edge of gradation change significantly.

Abstract:Nowadays, sentiment analysis has become a hot research topic in the natural language processing field. The automated and semi-supervised way of text sentiment analysis makes a high value on practicing and theory studies. The sentiment orientation algorithm based on sentiment lexicon is an important approach in text sentiment analysis. Constructing a sentiment lexicon effectively is a basic task in the text sentiment analysis. However, Chinese words are very ambiguous in different domains. Meanwhile, different areas of sentiment words also have the characteristic of specialized. To solve these problems, we propose a semi-supervised sentiment orientation classification algorithm based on word vector similarity (SO-WV). Experiments show that, the algorithm can classify the sentiment orientation of words effectively. This algorithm has the versatility in different areas, and also offers professional and specialized characteristics.

Abstract:This paper proposes an improved frequency estimation algorithm using discrete Fourier transform (DFT) interpolation based on the Quinn algorithm and iterative interpolation algorithm (A&M algorithm). The proposed algorithm first uses a frequency error estimated by the Quinn algorithm as the initial error value of the iterative estimation algorithm. Then frequency error is estimated accurately by the iteration algorithm. The algorithm can effectively reduce the number of iterations and guarantee the precision of estimation results, thus improving the computational efficiency. To enhance the efficiency of the algorithm on the DSP processor, this paper also proposes an optimization method for the implementation of the algorithm on the DSP processor, which is helpful for the application of the algorithm in real time. The simulation results show that the proposed algorithm can increase the frequency estimation accuracy, and the efficiency of real-time computation with good anti-noise performance.

Abstract:Although 2D-based face recognition technology becomes more and more mature, recognition results are still affected by light, posture, facial expressions and other changes. It is a trend to improve the performance of face recognition by 3D face model as well as to apply 3D face recognition in practice. To tackle these problems, SWJTU multimodal face database which contains face data from 200 Chinese people with neutral expression is proposed. The database includes visible light images, video sequences, 3D face models (high resolution) and stereo video sequences. Here, we describe the apparatuses, environments and procedure of the data collection and present the normalization procedure of the database. Finally, database applications are discussed and then several evaluation protocols for SWJTU multimodal face database are presented to measure face recognition and reconstruction performance.

Abstract:Entity linking includes entity discovery， query expansion， candidate generation， feature extraction and ranking. Here the query expansion method based on word embedding is proposed. Word embedding of words are trained by continuous bag-of-words (CBOW) model. Then the related words become the expansion words. The related words could make up the expansion based on rule. The related words could recall more and more candidate words simultaneously. In the feature extraction，the topic similarity between texts is extracted as the feature based on latent Dirichlet allocation(LDA). This paper extracts the synonyms based on word embedding as the dimension of text vector. Finally, learning to rank model is used to select the best candidate entity. The result shows that the method can ensure F1 reaching 0.71, and be effective for entity linking.

Abstract:Traditional signature-based method fails to identify the obfuscated malicious codes, while the dynamic method consumes a large amount of resources. Currently, most machine-learning-based detection methods cannot effectively distinguish trojan horses, worms and other malwares. Hence, we propose a new classification method based on malicious behavior features. The new method first learns specific malicious behavior sequential pattern of each malware category on the basis of the extraction of maliciousoriented instruction. The sample is projected to the new space which is composed of sequential patterns. Based on the new feature representation, a nearest neighbor classifier is constructed to classify the malicious codes. Experimental results show that the proposed method can effectively capture the malicious behavior and distinguish the differences among the behaviors of different malware categories, so as to improve the classification precision sharply.

Abstract:To reinforce the noise resistance capability of image edge detection algorithm better and refine edge information detection, a multi-scale morphology algorithm for image edge detection with noise resistance is proposed. Moreover, wavelet transform method is utilized to replace the commonly used weighted mean method and the edge images are obtained by each scale fused. The low frequency and high frequency coefficients of the wavelet decomposition are adopted respectively by different fusion strategies. Thus the edge details are effectively preserved. The fused image is clear and has rich details. Anti-noise detection algorithm is used to detect the edge of image using different scales of structuring elements. Hence, the algorithm becomes robust to the noise. The simulation result shows that the proposed algorithm can effectively reduce the influence of noise on detection results obtaining the ideal edge image.

Abstract:A method of sentiment analysis for online comment texts is proposd based on the latent Dirichlet allocation (LDA) model. The method extracts the sentiment information containing sentiment words and context with the sentiment word dictionary according to the specified collocation patterns of sentiment unit. Use the LDA model to mine the key features of the sentiment information and then combine them into the sentiment vector space. The machine-learning algorithm is used to classify the sentiment polarity of Chinese comment texts. After experiment, the presented method is proved to be effective in reducing dimensionality and text sentiment classification.

Abstract:The Vietnamese lexical features are discussed and essential characteristics of Vietnamese are integrated into condition random fields (CRFs) to propose a Vietnamese word segmentation method based on CRFs and ambiguity model. The segmentation corpus consisting of 25 981 Vietnamese is obtained as a training corpus of CRFs by computer marking and artificial proofreading. Vietnamese crossing ambiguity is widely distributed in the sentence. To eliminate the effects of crossing ambiguity, 5 377 ambiguity fragments are extracted from training corpus through dictionary of the forward and reverse matching algorithm. An ambiguity model is obtained by training the maximum entropy model. Then they are both incorparted into the segmentation model. The training corpus is divided into ten copies evenly for cross validation experiments. The segmentation accuracy reaches 96.55% in the experiment. Experimental results show that the method improves the segmentation accuracy rate, the recall rate and the F value of Vietnamese word obviously, compared with Vietnamese segmentation tool VnTokenizer.

Abstract:Gesture recognition based on small samples is one of the main trends in the advanced human-computer interaction research. A novel gesture recognition method based on adaptive K-nearest neighbor (A-KNN) and linear discriminant analysis (LDA) is presented. First, hand-shape images are segmented from the given interaction videos, and scaled to the same size to construct the training set. Then an optimized LDA algorithm is designed to extract gesture features. Finally, an improved KNN algorithm is introduced with adaptive Kvalue to classify the real-time gesture information. Test results show that the correct recognition rate of the proposed approach is higher than most existing methods. 