Vehicles equipped with 5th Generation(5G) wireless communication devices can exchange information with infrastructure(Vehicle to Infrastructure, V2I) to improve positioning accuracy. Vehicle location has great research value due to the problems of multipath environment and lack of Global Navigation Satellite System(GNSS) signals. This paper proposes a multi-frequency ranging method and positioning algorithm for 5G Orthogonal Frequency Division Multiplexing(OFDM) communication system. It selects specific subcarriers in the OFDM communication system to be used for transmitting ranging frames and delay observations without affecting other subcarriers used for communication. With almost no impact on communication capacity, several specific subcarriers of OFDM are used for ranging and positioning. It introduces the ranging subcarriers’ selection method and the format of the ranging frame carried by the subcarriers. The Cramero Lower Bound(CRLB) of this ranging positioning system is proved. Ranging positioning accuracy meets the requirements of vehicle location applications. The experimental simulation compares the performance with other positioning methods and proves the superiority of this system. The theory proves and simulates the relationship between ranging accuracy and channel parameters in a multipath environment. The simulation results show that the positioning accuracy about 5 cm can be achieved under the conditions of 5 GHz frequency and high signal-to-noise ratio(SNR).

In the Industrial Internet of Things (IIoT), various applications generate a large number of interactions and are vulnerable to various attacks, which are difficult to be monitored in a sophisticated way by traditional network architectures. Therefore, deploying software-defined networks (sdn) in the industrial IoT is essential to defend against various attacks. However, sdn has a drawback: there is a security problem of distributed denial-of-service attacks (DDoS) at the control layer. This paper proposes an effective solution: DDoS detection within the domain using tri-entropy in information theory. The detected attacks are then uploaded to a smart contract in the blockchain, so that the attacks can be quickly cut off even if the same attack occurs in different domains. Experimental validation was conducted under different attack strengths and multiple identical attacks, and the results show that the method has better detection ability under different attack strengths and can quickly block the same attacks.

Surface defect recognition on Train wheelset is crucial for the safe operation of a train wheel system. However, existing algorithms find it difficult to make rapid and accurate recognitions, due to the diversity and complexity of such defects. Targeting this issue, a train wheelset tread defect recognition method based on an improved spiking neural network (ISNN) is proposed. Specifically, a hybrid convolutional encoding module is first designed to conduct image-to-spike conversions and to create multi-scaled sparse representations of the features. Second, a residual spiking convolutional neural network is implemented to extract spiking features optimally, and a multi-scale structure is adopted to enhance the SNN’s ability to handle the details. A channel attention module is then incorporated to re-calibrate the weights of four-dimensional spiking feature maps. Finally, effective spiking features are obtained based on which the recognition decisions is made. The experimental results showed that the proposed method improved the accuracy of defect recognition. The recognition time of a single image is only 0.0195 s on average. The overall performance of the proposed method is noticeably superior to current mainstream algorithms.

Recommendation algorithms, from the perspective of real-time, can be classified as offline recommendation algorithms and online recommendation algorithms. To improve music recommendation accuracy, especially the new music (users have no historic listening records on it) recommendation accuracy, and real-time recommendation ability, and solve the “interest drift” problem, we propose a hybrid music recommendation model (HMRM) based on personalized measurement and game theory, which can be separated into two parts: an offline recommendation part (OFFLRP) and an online recommendation part (ONLRP). In OFFLRP, we emphasize users personalization. We introduce two metrics: User-Pursue-Novel-Degree (UPND) and Music Popularity (MP) to improve the traditional items-based collaborative filtering algorithm. In ONLRP, we try to solve the “interest drift” problem, which is a thorny problem in OFFLRP. We propose a novel online recommendation algorithm based on game theory. Experiments verify that the hybrid music recommendation model has a higher new music recommendation accuracy, a decent dynamical personalized recommendation ability, and real-time recommendation capability, and substantially mitigating the problem of interest drift.

A two-element dual-band flexible MIMO antenna which can be used for wearable applications is proposed in this paper. The antenna consists of two radiating elements fed by coplanar waveguide (CPW), and a shielding layer, which are all made of flexible conductive cloth MKKTN260. Each radiating element composes of two coupled split ring-shaped bending strips. The proposed antenna shows two measured impedance bandwidth (S11 < −10 dB) of 2.39-2.48 GHz and 5.72-5.88 GHz, so that it can be used for 2.4 GHz and 5.8 GHz ISM applications. The two coupled split rings form a self-isolation enhancement structure and can realize polarization diversity at 2.4 GHz band and radiation shielding at 5.8 GHz band, respectively. High isolation (>30 dB) has been achieved for both the bands. Other characteristics for wearable applications like gain, efficiency, SAR, and bending performances were also studied.

The spectrum resources are scarce to support the increasing throughput demands in vehicular networks. It is urgent to make full use of spectrum bands in mobile network. To get the availability of spectrum bands, users should sense wireless channels and cooperate with others. The spectrum sensing data are always related to users’ privacy, such as location. In this paper, we first introduce sensing trajectory inference attack (STIA) in cognitive vehicular networks (CVN), and then propose a data confusion-based privacy-preserving algorithm (DCPPA) and a cryptonym array-based privacy-preserving aggregation (CAPPA) scheme for spectrum sensing in CVN. Unlike existing methods, the proposed schemes transmit confused data during aggregation process. It is almost impossible to infer users’ location from the data transmitted. Analysis demonstrates that the proposed scheme are resilient to STIA

Clustering by fast search and find of density peaks (CFSFDP) has the advantages of a novel idea, easy implementation, and efficient clustering. It has been widely recognized in various fields since it was proposed in Science in 2014. The CFSFDP algorithm also has certain limitations, such as non-unified sample density metrics defined by cutoff distance, the “Domino Effect” for the assignment of remaining samples triggered by unstable assignment strategy, and the phenomenon of picking wrong density peaks as cluster centers. We propose reverse-nearest-neighbor-based clustering by fast search and find of density peaks (RNN-CFSFDP) to avoid these shortcomings. We redesign and unify the sample density metric by introducing reverse nearest neighbor. The newly defined local density metric and the K-nearest neighbors of each sample are combined to make the assignment process more robust and alleviate the “Domino Effect”. Specifically, a cluster fusion algorithm is proposed, which further alleviates the “Domino Effect” and effectively avoids the phenomenon of picking wrong density peaks as cluster centers. Experimental results on publicly available synthetic data sets and real-world data sets show that in most cases, the proposed algorithm is superior to or at least equivalent to the comparative methods in clustering performance. Especially, the proposed algorithm works better on manifold data sets and uneven density data sets.

In Unsupervised Video Object Segmentation (UVOS), the whole video might segment the wrong target due to the initial prior is missing. Also, in Semi-supervised Video Object Segmentation (SVOS), the initial video frame with a fine-grained pixel-level mask is essential to good segmentation accuracy. However, it is expensive and laborious to provide the accurate pixel-level masks for each training sequence. To address this issue, in this paper, we present a weak user interactive UVOS approach guided by a simple human-made rectangle annotation in the initial frame. Specifically, we first interactively draw the region of interest by a rectangle, and then we leverage the Mask RCNN method to generate a set of coarse reference labels for subsequent mask propagations. Second, to establish the temporal correspondence between the coherent frames, we further design two novel temporal modulation modules to enhance the target representations. Then, we compute the earth mover’s distance (EMD)-based similarity between coherent frames to mine the co-occurrent objects in the two images, which is used to modulate the target representation to highlight the foreground target. Furthermore, we design a cross-squeeze temporal modulation module to emphasize the co-occurrent features across frames, which further helps to enhance the foreground target representation. Finally, we augment the temporally modulated representations with the original representation and obtain the compositive spatio-temporal information, producing a more accurate VOS model. The experimental results on both UVOS and SVOS datasets including Davis2016, FBMS, Youtube-VOS, and Davis2017, show that the presented approach yields good balance between accuracy and complexity against the state-of-the-art solutions. The related code will be released at

.

In this paper, a Backward wave oscillator (BWO) with parallel multiple beams and multi-pin Slow-wave structure (SWS) operating at the frequency above 500 GHz is studied. Both the cold-cavity dispersion characteristics and CST Particle Studio (PIC) simulation results reveal that there are obvious mode competition problems in this kind of terahertz source. Considering that the structure of the multi-pin SWS is similar to that of two-dimensional Photonic crystals (PC), we introduce the defects of photonic crystal with the property of filtering into the SWS to suppress high-order modes. Furthermore, a detailed study of the effect of suppressing higher-order modes is carried out in the process of changing location and arrangement pattern of the point defects. The stable, single-mode operation of the terahertz source is realized. The simulation results show that the ratio of the output peak power of the higher-order modes to that of the fundamental mode is less than 1.9%. Also, the source can provide the output peak power of 44.8 mW at the frequency of 502.2 GHz in the case of low beam voltage of 4.7 kV.

As an extension of PE, coded permutation entropy (CPE) improves the performance of PE by making a secondary division for ordinal patterns defined in PE. In this study, we provide an exploration of the statistical properties of CPE using a finite length Gaussian white noise time series theoretically. By means of the Taylor series expansion, the approximate expressions of the expected value and variance of CPE are deduced and the Cramér-Rao Low Bound (CRLB) is obtained to evaluate the performance of the CPE estimator. The results indicate that CPE is a biased estimator, but the bias only depends on relevant parameters of CPE and it can be easily corrected for an arbitrary time series. The variance of CPE is related to the encoding patterns distribution, and the value converges to the CRLB of the CPE estimator when the time series length is large enough. For a finite-length Gaussian white noise time series model, the predicted values can match well with the actual values, which further validates the statistic theory of CPE. Using the theoretical expressions of CPE, it is possible to better understand the behavior of CPE for most of the time series.

With the development of the mobile communication and intelligent information technologies, the intelligent transportation systems (ITS) driven by the sixth generation (6G) has many opportunities to achieve ultra-low latency and higher data transmission rate. Nonetheless, it also faces the great challenges of spectral resource shortage and large-scale connection. To solve the above problems, non-orthogonal multiple access (NOMA) and cognitive radio (CR) technologies have been proposed. In this regard, we study the reliable and ergodic performance of CR-NOMA assisted ITS networks in the presence of imperfect successive interference cancellation (ipSIC) and non-ideal channel state information (CSI). Specifically, the analytical expressions of the outage probability (OP) and ergodic sum rate (ESR) are derived through a string of calculations. In order to gain more insights, the asymptotic expressions for OP and ESR at high signal-to-noise ratio (SNR) regimes are discussed. We verify the accuracy of the analysis by Monte Carlo simulations, and the results show: i) ipSIC and channel estimation errors (CEEs) have negative impacts on the OP and ESR; ii) The OP decreases with the SNR increasing until convergence to a fixed constant at high SNR regions; iii) The ESR increases with increasing SNR and there exists a ceiling in the high SNR region.

In federated learning, a parameter server may actively infer sensitive data of users and a user may arbitrarily drop out of a learning process. Bonawitz et al. propose a secure aggregation protocol for federated learning against a semi-honest adversary and a security enhancement method against an active adversary. The purpose of this paper is to analyze their security enhancement method and to design an alternative. We point out that their security enhancement method has the risk of Eclipse attack and that the consistency check round in their method could be removed. We give an new efficient security enhancement method by redesigning an authentication message and by adjusting the authentication timing. The new method produces an aggregation protocol secure against an active adversary with less communication and computation costs.

Based on an arbitrarily-chosen binary Golay complementary pair (BGCP)

of even length

${\boldsymbol{N}}$

, first of all, construct quadriphase sequences

$ {\boldsymbol{a}} $

and

$ {\boldsymbol{b}} $

of length

${\boldsymbol{N}}$

by weighting addition and difference of the aforementioned pair with different weights, respectively. Secondly, new quadriphase sequence

$ {\boldsymbol{u}} $

is given by interleaving three sequences

$ {\boldsymbol{d}} $

,

$ {\boldsymbol{a}} $

, and

$ -{\boldsymbol{c}} $

, and similarly, the sequence

$ {\boldsymbol{v}} $

is acquired from three sequences

$ {\boldsymbol{d}} $

,

$ {\boldsymbol{b}} $

, and

$ {\boldsymbol{c}} $

. Thus, the resultant pair

$ ({\boldsymbol{u}},{\boldsymbol{v}}) $

is the quadriphase Golay complementary pair (QGCP) of length

${\boldsymbol{3N}}$

. The QGCPs play a fairly important role in communications, radar, and so on.

As one of the most important railway signaling equipment, railway point machines undertake the major task of ensuring train operation safety. Thus fault diagnosis for railway point machines becomes a hot topic. Considering the advantage of the anti-interference characteristics of vibration signals, this paper proposes an novel intelligent fault diagnosis method for railway point machines based on vibration signals. A feature extraction method combining variational mode decomposition (VMD) and multiscale fluctuation-based dispersion entropy (MFDE) is developed, which is verified a more effective tool for feature selection. Then, a two-stage feature selection method based on Fisher discrimination and ReliefF is proposed, which is validated more powerful than signle feature selection methods. Finally, support vector machine (SVM) is utilized for fault diagnosis. Experiment comparisons show that the proposed method performs best. The diagnosis accuracies of normal-reverse and reverse-normal switching processes reach 100% and 96.57% respectively. Especially, it is a try to use new means for fault diagnosis on railway point machines, which can also provide references for similar fields.

The development of smart mobile devices not only brings convenience to people’s lives but also provides a breeding ground for Android malware. The sharp increasing malware poses a disastrous threat to personal privacy in the information age. Based on the fact that malware heavily resorts to system APIs to perform its malicious actions, there has been a variety of API-based detection approaches. Most of them do not consider the relationship between APIs. We contribute a new approach based on Enhanced API Order for Android malware detection, named EAODroid. EAODroid learns the similarity of system APIs from a large number of API sequences and groups similar APIs into clusters. The extracted API clusters are further used to enhance the original API calls executed by an app to characterize behaviors and perform classification. We perform multi-dimensional experiments to evaluate EAODroid on three datasets with ground truth. We compare with many state-of-the-art works, showing that EAODroid achieves effective performance in Android malware detection.

A novel Wilkinson pulse combiner(WPC) is proposed for the combination of Gaussian pulse signals. The WPC requires a very wide bandwidth, small size and high port isolation. To improve the operating bandwidth, the design adopts the form of eight-section WPC. Eight capacitors are connected in series with the isolating resistors of each section. After capacitive loading, isolation between WPC input ports is significantly improved at low frequency. Consequently, the operating bandwidth of WPC has been increased from 13:1 to 31:1. Compared with the conventional Wilkinson combiner with the same bandwidth, the proposed WPC reduces the size by 40%. In addition, all the ports are well impedance matched and the insertion loss in the operating frequency band is less than 0.5dB. To verify the feasibility of the design, a prototype was fabricated and measured. Experiment shows that the novel WPC is more advantageous to generate dual-Gaussian pulse signals.

Parallel simulation generally needs efficient, reliable and order-preserving communication. In this article, a zero-copy, reliable and order-preserving intra-node message passing approach ZeROshm is proposed, and it partitions shared memory into segments assigned to processes for receiving messages. Each segment consists of two levels of index L1 and L2 that recordes the order of messages in the host segment, and the processes also read from and write to the segments directly according to the indexes, thereby eliminating allocating and copying buffers. As experimental results show, ZeROshm exhibits nearly equivalent performance to MPI for small message and superior performance for large message - ZeROshm costs less time by 43%, 40% and 55% respectively in pure communication, communication with contention and real Phold simulation within a single node. In hybrid environment, the combination of ZeROshm and MPI also shorten the execution time of Phold simulation by about 42% compared to pure MPI.

The strength of any cryptographic algorithm is mostly based on the difficulty of its encryption key.However, the larger size of the shared key the more computational operations and processing time for cryptographic algorithms. To avoid increasing the key size and keep its secrecy, we must hide it. The authors proposed a stream cipher algorithm that can hide the symmetric key^[1] through hashing and splitting techniques. This paper aims to measure security analysis and performance assessment for this algorithm. This algorithm is compared with three of the commonly used stream cipher algorithms: RC4, Rabbit, and Salsa20 in terms of execution time and throughput. This comparison has been conducted with different data types as audio, image, text, docs, and pdf. Experiments proved the superiority of SCKHA algorithm over both Salsa20 and Rabbit algorithms. Also, results proved the difficulty to recover the secret key for SCKHA algorithm. Although RC4 has a lower encryption time than SCKHA, it is not recommended for use because of its vulnerabilities. Security factors that affect the performance as avalanche effect, correlation analysis, histogram analysis, and Shannon information entropy are highlighted. Also, the ciphertext format of the algorithm gives it the ability to search over encrypted data.

Unmanned aerial vehicles (UAVs) can be effectively used as serving stations in emergency communications because of their free movements, strong flexibility, and dynamic coverage. In this paper, we propose a coordinated multiple points (CoMP) based UAV deployment framework to improve system average ergodic rate, by using the fuzzy C-means (FCM) algorithm to cluster the ground users and considering exclusive forest channel models for the two cases, i.e., associated with a broken base station (BS) or an available one. In addition, we derive the upper bound of the average ergodic rate to reduce computational complexity. Since deep reinforcement learning (DRL) can deal with the complex forest environment while the large action and state space of UAVs leads to slow convergence, we use a ratio cut method to divide UAVs into groups and propose a hierarchical clustering DRL (HC-DRL) approach with quick convergence to optimize the UAV deployment. Simulation results show that the proposed framework can effectively reduce the complexity, and outperforms the counterparts in accelerating the convergence speed.

Compared with the traditional short-term object tracking task based on temporal-spatial consistency, the long-term object tracking task faces the challenges of object disappearance, dramatic changes in object scale, and object appearance. To address these challenges and problems, in this paper we propose a Multi-scale Global Retrieval and Temporal-Spatial Consistency Matching based long-term Tracking Network (MTTNet). MTTNet regards the long-term tracking task as a single sample object detection task and takes full advantage of the temporal-spatial consistency assumption between adjacent video frames to improve the tracking accuracy. MTTNet utilizes the information of single sample as guidance to perform full-image multi-scale retrieval on any instance and does not require online learning and trajectory refinement. Any type of error generated during the detection process will not affect its performance on subsequent video frames. This can overcome the accumulation of errors in the tracking process of traditional object tracking networks. We introduce Atrous Spatial Pyramid Pooling to address the challenge of dramatic changes in the scale and the appearance of the object. On the experimental results, MTTNet can achieve better performance than composite processing methods on two large datasets.

In this paper, intelligent reflecting surface (IRS) is introduced to enhance the performance of cognitive radio (CR) systems. The robust beamforming is designed based on combined bounded channel state information (CSI) error for primary user (PU) related channels. The transmit precoding at the secondary user (SU) transmitter and phase shifts at the IRS are jointly optimized to minimize the SU's total transmit power subject to the quality of service of SUs, the limited interference imposed on the PU and unit-modulus of the reflective beamforming. Simulation results verify the efficiency of the proposed algorithm and reveal that the number of phase shifts at IRS should be carefully chosen to obtain a tradeoff between the total minimum transmit power and the feasibility rate of the optimization problem.

There are many kinds of linear canonical transform (LCT)-based Wigner distributions (WDs), which are both very effective in detecting noisy linear frequency-modulated (LFM) signals. Among WDs in LCT domains, the instantaneous cross-correlation function type of Wigner distribution (ICFWD) attracts much attention from scholars, because it achieves not only low computational complexity but also good detection performance. However, the existing LCT free parameters selection strategy, a solution of the expectation-based output signal-to-noise ratio (SNR) optimization model, is not unique. In this paper, by introducing the variance-based output SNR optimization model, a multiobjective optimization model is established. Then the existence and uniqueness of the optimal parameters of ICFWD are investigated. The solution of the multiobjective optimization model with respect to one-component LFM signal added with zero-mean stationary circular Gaussian noise is derived. A comparison of the unique parameters selection strategy and the previous one is carried out. The theoretical results are also verified by numerical simulations.

Phase filtering is an essential step in interferometric synthetic aperture radar (InSAR). For interferograms with complicated and changeable terrain, the increasing resolution of InSAR images makes it even more difficult. In this paper, a novel adaptive InSAR phase filtering method based on complexity factors is proposed. Firstly, three complexity factors based on the noise distribution and terrain slope information of the interferogram are selected. The complexity indicator composed of three complexity factors is used to guide the adaptive selection of the most suitable and effective filtering strategies for different areas. Then, the complexity scalar is calculated, which can guide the adaptive local fringe frequency (LFF) estimation and adaptive parameters calculation in different filter methods. Finally, validations are performed on the simulated and real data. The performance comparison between the other three representative phase filtering method and the proposed method have validated the effectiveness and superiority of the proposed method.

In this paper, we propose a terahertz radiation source over the R-band (220-325 GHz) based on metamaterials (MTMs) structure and parallel multiple beams. The effective permittivity and permeability of the slow-wave structure (SWS) can be obtained through the S-parameter retrieval approach, using numerical simulation. Additionally, the electromagnetic properties of the MTMs structure are analyzed, including the dispersion and the coupling impedance. Furthermore, we simulate the beam-wave interaction of the backward oscillator (BWO) with MTMs structure and parallel multiple beams by 3-D particle-in-cell (PIC) code. It is observed that parallel multiple beams can highly enhance the beam-wave interaction and greatly enlarge the output power. These results indicate that the saturated (peak) output power is approximately 63W with the efficiency of roughly 6% at the operating frequency of 231 GHz, under the beam voltage of 35 kV and total current of 30 mA (6-beam) respectively. Meanwhile, the BWO can generate power of 10 W-80 W in the tunable frequency of 220 GHz-240 GHz.

In this paper, a 3.6 mm gate width GaN HEMT with 0.35 μm gate length process and input and output matching circuits of Nanjing Electronic Devices Institute are used for broadband design respectively, and a novel high-power and high-efficiency self-bias S-band broadband continuous wave GaN power amplifier is realized. Under the working conditions of 2.2 GHz to 2.6 GHz and 32 V drain power supply, the continuous wave output power of the amplifier is more than 20 W, the power gain is more than 15 dB, and the max power added efficiency is more than 65%. The self-bias amplifier simplifies the circuit structure and realizes excellent circuit performance.

Feature-based (FB) algorithms are widely used in modulation classification due to their low complexity. As a prerequisite step of FB, feature selection can reduce the computational complexity without significant performance loss. In this paper, according to the linear separability of cumulant features, the hyperplane of the support vector machine is used to classify modulation types, and the contribution of different features is ranked through the weight vector. Then, cumulant features are selected using recursive feature elimination (RFE) to identify the modulation type employed at the transmitter. We compare the performance of the proposed algorithm with existing feature selection algorithms and analyze the complexity of all the mentioned algorithms. Simulation results verify that the proposed RFE algorithm can optimize the selection of the features to realize modulation recognition and improve identification efficiency.

Soluble sugar is an important index to determine the quality of jujube, and also an important factor to influence the taste of jujube. The soluble sugar content of jujube mainly depends on manual chemical measurement, which is time-consuming and labor-intensive. In this study, the feasibility of multi-spectral imaging combined with deep learning for rapid nondestructive testing of fruit internal quality was analyzed. Support vector machine regression model, partial least squares regression model and convolutional neural networks (CNNs) model were established by multi-spectral imaging method to predict the soluble sugar content of the whole jujube fruit, and the optimal model was selected to predict the content of three kinds of soluble sugar. The study showed that the sucrose prediction model of the whole jujube had the best performance after CNNs training, and the correlation coefficient of verification set was 0.88, which proved the feasibility of using CNNs for prediction of the soluble sugar content of jujube fruits.

In this article, a design approach to a radial-resonant wide beamwidth circular sector patch antenna is advanced. As properly evolved from a U-shaped dipole, a prototype magnetic dipole can be fit in the radial direction of a circular sector patch radiator, with its length set as the positive odd-integer multiples of one-quarter wavelength. In this way, multiple TM_0m (m = 1, 2, …) modes resonant circular sector patch antenna with short-circuited circumference and widened E-plane beamwidth can be realized by proper excitation and perturbations. Prototype antennas are then designed and fabricated to validate the design approach. Experimental results reveal that the E-plane beamwidth of a dual-resonant antenna fabricated on air/Teflon substrate can be effectively broadened to 128°/120°, with an impedance bandwidth of 17.4%/7.1%, respectively. In both cases, the antenna heights are strictly limited to no more than 0.03-guided wavelength. It is evidently validated that the proposed approach can effectively enhance the operational bandwidth and beamwidth of a microstrip patch antenna while maintaining its inherent low profile merit.

Speech keyword spotting system is a critical component of human-computer interfaces. And Connectionist temporal classifier (CTC) has been proven to be an effective tool for that task. However, the standard training process of speech keyword spotting faces a data imbalance issue where positive samples are usually far less than negative samples. Numerous easy-training negative examples overwhelm the training, resulting in a degenerated model. To deal with it, this paper tries to reshape the standard CTC loss and proposes a novel re-weighted CTC loss. It evaluates the sample importance by its number of detection errors during training and automatically down-weights the contribution of easy examples, the majorities of which are negatives, making the training focus on samples deserving more training. The proposed method can alleviate the imbalance naturally and make use of all available data efficiently. Evaluation on several sets of keywords selected from AISHELL-1 and AISHELL-2 achieves 16%—38% relative reductions in false rejection rates over standard CTC loss at 0.5 false alarms per keyword per hour in experiments.

We consider the general problem of the approximate joint diagonalization of a set of non-Hermitian matrices. This problem mainly arises in the data model of the joint blind source separation for two datasets. Based on a special parameterization of the two diagonalizing matrices and on adapted approximations of the classical cost function, we establish two Jacobi-like algorithms. They may serve for the canonical polyadic decomposition (CPD) of a third-order tensor, and in some scenarios they can outperform traditional CPD methods. Simulation results demonstrate the competitive performance of the proposed algorithms.

Constrained by physics, the spatial resolution of hyperspectral images (HSI) is low. Hyperspectral image super-resolution (HSI SR) is a task to obtain high-resolution hyperspectral images (HR HSI) from low-resolution hyperspectral images (LR HSI). Existing algorithms have the problem of losing important spectral information while improving spatial resolution. To handle this problem, a spatial-spectral feature extraction network (SSFEN) for HSI SR is proposed in this paper. It enhances the spatial resolution of the HSI while preserving the spectral information. The SSFEN is composed of three parts: spatial-spectral mapping network (SSMN), spatial reconstruction network (SRN), and spatial-spectral fusing network (SSFN). And a joint loss function with spatial and spectral constraints is designed to guide the training of the SSFEN. Experiment results show that the proposed method improves the spatial resolution of the HSI and effectively preserves the spectral information coinstantaneously.

Essential proteins with biological functions are necessary for the survival of organisms. Computational recognition methods of essential proteins can reduce the workload and provide candidate proteins for biologists. However, existing methods fail to efficiently identify essential proteins, and generally do not fully use amino acid sequence information to improve the performance of essential protein recognition. In this work, we proposed an end-to-end deep contextual representation learning framework called DeepIEP to automatically learn biological discriminative features without prior knowledge based on protein network heterogeneous information. Specifically, the model attaches amino acid sequences as the attributes of each protein node in the protein interaction network, and then automatically learns topological features from protein interaction networks by graph embedding algorithms. Next, multi-scale convolutions and gated recurrent unit networks are used to extract contextual features from gene expression profiles. The extensive experiments confirm that our DeepIEP is an effective and efficient feature learning framework for identifying essential proteins and contextual features of protein sequences can improve the recognition performance of essential proteins.

With the recent advances in the field of deep learning, an increasing number of deep neural networks have been applied to business process prediction tasks, remaining time prediction, to obtain more accurate predictive results. However, existing time prediction methods based on deep learning have poor interpretability, an explainable business process remaining time prediction method is proposed using reachability graph, which consists of prediction model construction and visualization. For prediction models, a Petri net is mined and the reachability graph is constructed to obtain the transition occurrence vector. Then, prefixes and corresponding suffixes are generated to cluster into different transition partitions according to transition occurrence vector. Next, the bidirectional recurrent neural network with attention is applied to each transition partition to encode the (trace) prefixes, and the deep transfer learning between different transition partitions is performed. For the visualization of prediction models, the evaluation values are added to the sub-processes of Petri net to realize the visualization of the prediction models. Finally, the proposed method is validated by publicly available event logs.

Weakly supervised semantic segmentation using just image-level labels is critical since it alleviates the need for expensive pixel-level labels. Most cutting-edge methods adopt two-step solutions that learn to produce pseudo-ground-truth using only image-level labels and then train off-the-shelf fully supervised semantic segmentation network with these pseudo labels. Although these methods have made significant progress, they also increase the complexity of the model and training. In this paper, we propose a one-step approach for weakly supervised image semantic segmentation—Attention guided enhancement network (AGEN), which produces pseudo-pixel-level labels under the supervision of image-level labels and trains the network to generate segmentation masks in an end-to-end manner. Particularly, we employ Class activation maps (CAM) produced by different layers of the classification branch to guide the segmentation branch to learn spatial and semantic information. However, the CAM produced by the lower layer can capture the complete object region but with many noises. Thus, the self-attention module is proposed to enhance object regions adaptively and suppress irrelevant object regions, further boosting the segmentation performance. Experiments on the Pascal VOC 2012 dataset show that the performance of AGEN outperforms other state-of-art weakly supervised semantic segmentation with only image-level labels.

The shuffle operations are the bottleneck when mapping the FFT-like algorithms to the vector SIMD architectures. We propose six (three pairs) innovative vector memory-access shuffle fused instructions,which have been proved mathematically. Together with the proposed modified binary-exchange method,the innovative instructions can efficiently address the bottleneck problem for DIF/DIT radix-2/4 FFT-like algorithms,reach a performance improvement by 17.9%~111.2% and reduce the code size by 5.4%~39.8%.Besides,the proposed instructions fit some hybrid-radix FFTs and are suitable for the terms of the initial or result data placement for general algorithms. The software and hardware cost of the proposed instructions is moderate.

Assuming unknown knowledge of Target impulse response (TIR), this paper deals with the joint design of Multiple-input multiple-output (MIMO) Space-time transmit code (STTC) and Space-time receive filter (STRF) for the detection of extended targets in the presence of signal-dependent interference. To enhance the detection performance of extended targets for MIMO radar, we consider transmit-receive system optimization to maximize the worst-case Signal to interference plus noise ratio (SINR) at the output of the STRF array. The problem is formulated in terms of a non-convex max-min quadratic fractional optimization program. Relying on an appropriate reformulation, we present an alternate optimization technique which monotonically increases the SINR value and converges to a stationary point. All iterations of the procedure, involve both a convex and a max-min quadratic fractional programming problem which is globally solved resorting to the generalized Dinkelbachos process with a polynomial computational complexity. In addition, resorting to several mathematical manipulations, the original problem is transformed into an equivalent convex problem, which can also be globally solved via interior-point methods. Finally, the effectiveness of two optimization design procedures is demonstrated through experimental results, underlining the performance enhancement offered by robust joint design methods.