To tackle the challenge of applying convolutional neural network (CNN) in field-programmable gate array (FPGA) due to its computational complexity, a high-performance CNN hardware accelerator based on Verilog hardware description language was designed, which utilizes a pipeline architecture with three parallel dimensions including input channels, output channels, and convolution kernels. Firstly, two multiply-and-accumulate (MAC) operations were packed into one digital signal processing (DSP) block of FPGA to double the computation rate of the CNN accelerator. Secondly, strategies of feature map block partitioning and special memory arrangement were proposed to optimize the total amount of off-chip access memory and reduce the pressure on FPGA bandwidth. Finally, an efficient computational array combining multiplicative-additive tree and Winograd fast convolution algorithm was designed to balance hardware resource consumption and computational performance. The high parallel CNN accelerator was deployed in ZU3EG of Alinx, using the YOLOv3-tiny algorithm as the test object. The average computing performance of the CNN accelerator is 127.5 giga operations per second (GOPS). The experimental results show that the hardware architecture effectively improves the computational power of CNN and provides better performance compared with other existing schemes in terms of power consumption and the efficiency of DSPs and block random access memory (BRAMs).

In this paper, the performance of you only look once ( YOLO) series detectors on Chinese license plate recognition (LPR) in the real intelligent transportation system (ITS) monitoring scene is investigated. Specially, a precise and efficient automatic license plate recognition ( ALPR ) system based on the YOLOv4 detector is proposed. The proposed ALPR system contains three stages including vehicle detection, license plate detection (LPD) and LPR. In vehicle detection stage, YOLOv4 detector is directly applied. In LPD stage, YOLOv4-tiny detector is exploited. In the last stage, the YOLOv4-tiny detector with attention mechanism for LPR is proposed to use. In addition, a large Chinese license plate dataset containing 10 500 images collected from all 31 provinces in the Chinese mainland is created. This Chinese license plate dataset is named Hefei University of Technology license plate version 1 (HFUT-LP v1). Particularly, HFUT-LP v1 dataset is collected in the real ITS monitoring scene. In order to compare the performance of different object detection algorithms for ALPR, a variety of object detection algorithms are used to make a comprehensive performance evaluation. Experimental results show that the proposed ALPR system achieves very high accuracy and has very fast processing speed, which is suitable for real-time LPR.

As the traditional semiconductor complementary metal oxide semiconductor (CMOS) integrated circuit technology gradually approaches the limit of Moore's Law, quantum computing, as a new system computing technology,with the potential for higher computing speed and lower power consumption, is getting more and more attention from governments and research institutions around the world. For instance, the United States (US) government is adopting a bunch of bills to deploy new quantum information processing technology. The European Union's “quantum declaration" plans to realize customized quantum computers with more than 100 qubits in the next 10 years. The Chinese government also provides strong support for quantum computer research through the project of National Science and Technology Major Projects. In the business field, major companies such as International Business Machines (IBM) Corporation, Intel, Google, Alibaba, Huawei, Baidu, etc. , have joined the “quantum supremacy" competition in order to seize the initiative in the future information field. Facing the rapid development trend of quantum computing, we believe that we should refer to the classical computer industry in the early stage, form an industrial system, and develop the quantum computing industry. We should also use the scientific system engineering method to carry out the research and development of the quantum computer and establish the ecological environment of the quantum computer industry with the demand as the traction, so as to better serve the development of the national economy.

Mobile robots have been used for many industrial scenarios which can realize automated manufacturing process instead of human workers. To improve the quality of the optimal rapidly-exploring random tree ( RRT* ) for planning path in dynamic environment, a high-quality dynamic rapidly-exploring random tree ( HQD-RRT* ) algorithm is proposed in this paper, which generates a high-quality solution with optimal path length in dynamic environment. This method proceeds in two stages: initial path generation and path re-planning. Firstly, the initial path is generated by an improved smart rapidly-exploring random tree ( RRT* -SMART) algorithm, and the state tree information is stored as prior knowledge. During the process of path execution, a strategy of obstacle avoidance is proposed to avoid moving obstacles. The cost and smoothness of path are considered to re-plan the initial path to improve the path quality in this strategy. Compared with related work, a higher-quality path in dynamic

environment can be achieved in this paper. HQD-RRT* algorithm can obtain an optimal path with better stability. Simulations on the static and dynamic environment are conducted to clarify the efficiency of HQD-RRT* in avoiding unknown obstacles.

To deal with the secrecy issues and energy efficiency issues in the unmanned aerial vehicles ( UAVs) assisted communication systems, an UAV-enabled multi-hop mobile relay system is studied in an urban environment. Multiple rotary-wing UAVs with energy budget considerations are employed as relays to forward confidential information between two ground nodes in the presence of multiple passive eavesdroppers. The system secrecy energy efficiency ( SEE), defined by the ratio of minimum achievable secrecy rate ( SR) to total propulsion energy consumption (PEC), is maximized via jointly optimizing the trajectory and transmit power of each UAV relay. To solve the formulated non-convex fractional optimization problem subject to mobility, transmit power and information-causality constraints, an effective iterative algorithm is proposed by applying the updated-rate-assisted block coordinate decent method, successive convex approximation (SCA) technique and Dinkelbach method. Simulation

results demonstrate the effectiveness of the proposed joint trajectory design and power control scheme.

To solve polynomial systems, Harrow, Hassidim, and Lloyd (HHL) proposed a quantum algorithm called HHL algorithm. Based on the HHL algorithm, Chen et al. presented an algorithm, the solving the Boolean solutions of polynomial systems (PoSSoB) algorithm. Furthermore, Ding et al. introduced the Boolean Macaulay matrix and analyzed the lower bound on the condition number. Inspired by Ding et al. 's research, several related algorithms are proposed in this paper. First, the improved PoSSoB algorithm using the Boolean Macaulay matrix is proved to have lower complexity. Second, for solving equations with errors, a quantum algorithm for the max-polynomial system solving (Max-PoSSo) problem is proposed based on the improved PoSSoB algorithm. Besides, the Max-PoSSo algorithm is extended to the learning with errors (LWE) problem and its special case, the learning parity with noise (LPN) problem, providing a quantitative criterion, the condition number, for the security of these basic problems.

The task of multimodal sentiment classification aims to associate multimodal information, such as images and texts with appropriate sentiment polarities. There are various levels that can affect human sentiment in visual and textual modalities. However, most existing methods treat various levels of features independently without having effective method for feature fusion. In this paper, we propose a multi-level fusion classification (MFC) model to predict the sentiment polarity based on the fusing features from different levels by exploiting the dependency among them. The proposed architecture leverages convolutional neural networks ( CNNs) with multiple layers to extract levels of features in image and text modalities. Considering the dependencies within the low-level and high-level features, a bi-directional (Bi) recurrent neural network (RNN) is adopted to integrate the learned features from different layers in CNNs. In addition, a conflict detection module is incorporated to address the conflict between modalities. Experiments on the Flickr dataset demonstrate that the MFC method achieves comparable performance compared with strong baseline methods.

Soft margin support vector machine (SVM) with hinge loss function is an important classification algorithm, which has been widely used in image recognition, text classification and so on. However, solving soft margin SVM with hinge loss function generally entails the sub-gradient projection algorithm, which is very time-consuming when processing big training data set. To achieve it, an efficient quantum algorithm is proposed. Specifically, this algorithm implements the key task of the sub-gradient projection algorithm to obtain the classical sub-gradients in each iteration, which is mainly based on quantum amplitude estimation and amplification algorithm and the controlled rotation operator. Compared with its classical counterpart, this algorithm has a quadratic speedup on the number of training data points. It is worth emphasizing that the optimal model parameters obtained by this algorithm are in the classical form rather than in the quantum state form. This enables the algorithm to classify new data at little cost when the optimal model parameters are determined.

As the core technology of optical networks virtualization, virtual optical network embedding ( VONE) enables multiple virtual network requests to share substrate elastic optical network ( EON) resources simultaneously and hence has been applicated in edge computing scenarios. In this paper, we propose a reinforced virtual optical network embedding ( R-VONE ) algorithm based on deep reinforcement learning ( DRL) to optimize network embedding policies automatically. The network resource attributes are extracted as the environment state for model training, based on which DRL agent can deduce the node embedding probability. Experimental results indicate that R-VONE presents a significant advantage with lower blocking probability and higher resource utilization.

Adaptive learning paths provide individual learning objectives that best match a learner's characteristics. This is especially helpful when learners need to balance limited available learning time and multiple learning objectives. The automatic generation of personalized learning paths to improve learning efficiency has therefore attracted significant interest. However, most current research only focuses on providing learners with adaptive objects and sequences according to their own interests or learning goals given a normal amount of time or ordinary conditions. There is little research that can help learners to obtain the most important knowledge for a test in the shortest time possible, which is a typical scenario in exanimation-oriented education systems. This study aims to solve this problem by introducing a new approach that builds on existing methods. First, the eight properties in Gardner's multiple intelligence theory are introduced into the present knowledge and learner models to define the relationship between learning objects (LOs) and learners, thereby improving recommendation accuracy rates. Then, a novel adaptive learning path recommendation model is presented where viable knowledge topologies, knowledge bases and the previously-established properties relating to a learner's ability are combined by Dempster-Shafer (D-S) evidence theory. A series of practical experiments were performed to assess the approach's adaptability, the appropriateness of the selected evidence and the effectiveness of the recommendations. In the results, it was found that the proposed learning path recommendation model helped learners learn the most important elements and obtain superior test grades when confronted with limited time for learning.

It is a critical challenge for quantum machine learning to classify the datasets accurately. This article develops a quantum classifier based on the isolated quantum system (QC-IQS) to classify nonlinear and multidimensional datasets. First, a model of QC-IQS is presented by creating parameterized quantum circuits (PQCs) based on the decomposing of unitary operators with the Hamiltonian in the isolated quantum system. Then, a parameterized quantum classification algorithm (QCA) is designed to calculate the classification results by updating the loss function until it converges. Finally, the experiments on nonlinear random number datasets and Iris datasets are designed to demonstrate that the QC-IQS model can handle and generate accurate classification results on different kinds of datasets. The experimental results reveal that the QC-IQS is adaptive and learnable to handle different types of data. Moreover, QC-IQS compensates the issue that the accuracy of previous quantum classifiers declines when dealing with diverse datasets. It promotes the process of novel data processing with quantum machine learning and has the potential for more comprehensive applications in the future.

Fine-grained few-shot learning is a difficult task in image classification. The reason is that the discriminative
features of fine-grained images are often located in local areas of the image, while most of the existing few-shot learning image classification methods only use top-level features and adopt a single measure. In that way, the local features of the sample cannot be learned well. In response to this problem, ensemble relation network with multi-level measure (ERN-MM) is proposed in this paper. It adds the relation modules in the shallow feature space to compare the similarity between the samples in the local features, and finally integrates the similarity scores from the feature spaces to assign the label of the query samples. So the proposed method ERN-MM can use local details and global information of different grains. Experimental results on different fine-grained datasets show that the proposed method achieves good classification performance and also proves its rationality.

With the rapid development of Internet of thing (IoT) technology, it has become a challenge to deal with the increasing number and diverse requirements of IoT services. By combining burgeoning network function virtualization ( NFV) technology with cloud computing and mobile edge computing ( MEC), an NFV-enabled cloud-and-edge-collaborative IoT (CECIoT) architecture can efficiently provide flexible service for IoT traffic in the form of a service function chain (SFC) by jointly utilizing edge and cloud resources. In this promising architecture, a difficult issue is how to balance the consumption of resource and energy in SFC mapping. To overcome this challenge, an intelligent energy-and-resource-balanced SFC mapping scheme is designed in this paper. It takes the comprehensive deployment consumption as the optimization goal, and applies a deep Q-learning(DQL)-based SFC mapping (DQLBM) algorithm as well as an energy-based topology adjustment (EBTA) strategy to make efficient use of the limited network resources, while satisfying the delay requirement of users. Simulation results show that the proposed scheme can decrease service delay, as well as energy and resource consumption.

Personalized recommender systems provide various personalized recommendations for different users through the analysis of their respective historical data. Currently, the problem of the “filter bubble" which has to do with over-specialization persists. Serendipity (SRDP), one of the evaluation indicators, can provide users with unexpected and useful recommendations, and help to successfully mitigate the filter bubble problem, and enhance users' satisfaction levels and provide them with diverse recommendations. Since SRDP is highly subjective and challenging to study, only a few studies have focused on it in recent years. In this study, the research results on SRDP were summarized, the various definitions of SRDP and its applications were discussed, the specific SRDP calculation process from qualitative to quantitative perspectives was presented, the challenges and the development directions were outlined to provide a framework for further research.

This paper focuses on gesture recognition and interactive lighting control. The collection of gesture data adopts the Myo armband to obtain surface electromyography (sEMG). Considering that many factors affect sEMG, a customized classifier based on user calibration data is used for gesture recognition. In this paper, machine learning classifiers k-nearest neighbor (KNN), support vector machines (SVM), and naive Bayesian (NB) classifier, which can be used in small sample sets, are selected to classify four gesture actions. The performance of the three classifiers under different training parameters, different input features, including root mean square (RMS), mean absolute value (MAV), waveform length (WL), slope sign change (SSC) number, zero crossing (ZC) number, and variance (VAR) are tested, and different input channels are also tested. Experimental results show that: The NB classifier, which assumes that the prior probability of features is polynomial distribution, has the best performance, reaching more than 95% accuracy. Finally, an interactive stage lighting control system based on Myo armband gesture recognition is implemented.