Search

Modern real-world control problems call for continuous control domains and robust, sample efficient and explainable control frameworks. We are presenting a framework for recursively composing control skills to solve compositional and progressively complex tasks. The framework promotes reuse of skills, and as a result quick adaptability to new tasks. The decision tree can be observed, providing insight into the agents' behavior. Furthermore, the skills can be transferred, modified or trained independently, which can simplify reward shaping and increase training speeds considerably. This paper is concerned with efficient composition of control algorithms using reinforcement learning and soft attention. Compositional and temporal abstraction is the key to improving learning and planning in reinforcement learning. Our Thompson sampling inspired soft-attention model is demonstrated to efficiently solve the composition problem. [ABSTRACT FROM AUTHOR]

Harris Hawks optimization (HHO) algorithm was a powerful metaheuristic algorithm for solving complex problems. However, HHO could easily fall within the local minimum. In this paper, we proposed an improved Harris Hawks optimization (IHHO) algorithm for solving different engineering tasks. The proposed algorithm focused on random location-based habitats during the exploration phase and on strategies 1, 3, and 4 during the exploitation phase. The proposed modified Harris hawks in the wild would change their perch strategy and chasing pattern according to updates in both the exploration and exploitation phases. To avoid being stuck in a local solution, random values were generated using logarithms and exponentials to explore new regions more quickly and locations. To evaluate the performance of the proposed algorithm, IHHO was compared to other five recent algorithms [grey wolf optimization, BAT algorithm, teaching–learning-based optimization, moth-flame optimization, and whale optimization algorithm] as well as three other modifications of HHO (BHHO, LogHHO, and MHHO). These optimizers had been applied to different benchmarks, namely standard benchmarks, CEC2017, CEC2019, CEC2020, and other 52 standard benchmark functions. Moreover, six classical real-world engineering problems were tested against the IHHO to prove the efficiency of the proposed algorithm. The numerical results showed the superiority of the proposed algorithm IHHO against other algorithms, which was proved visually using different convergence curves. Friedman's mean rank statistical test was also inducted to calculate the rank of IHHO against other algorithms. The results of the Friedman test indicated that the proposed algorithm was ranked first as compared to the other algorithms as well as three other modifications of HHO. [ABSTRACT FROM AUTHOR]

Non-blind image deblurring has attracted a lot of attention in the field of low-level vision. However, the existing non-blind deblurring methods cannot effectively deal with a saturated blurry image. The key point is that the degradation model of saturated blurry images does not satisfy the linear convolution model of a conventional blurry image. To solve the problem, in this paper, we proposed a novel deep non-blind deblurring method, dubbed saturated image non-blind deblurring network(SDBNet). The SDBNet contains two trainable sub-network, i.e., confident estimate network (CEN) and detail enhance network (DEN). Specifically, the SDBNet uses CEN to estimate the confidence map for the saturated blurry image, which is used to recognize saturated pixels in the blurry image, and then uses the confidence map, and blur kernel to restore the blurry image. Finally, we use DEN to enhance the edges and textures of the restored image. We first pre-train CEN and DEN. In order to effectively pre-train CEN, we propose a new robust function, which is used to generate label data for CEN. The experimental results show that compared with several existing non-blind deblurring methods, SDBNet can effectively restore saturated blurry images and better restore the texture, edge, and other structural information of blurry images. [ABSTRACT FROM AUTHOR]

Intelligent swarm optimization algorithms have become increasingly common due to their success in solving real-world problems. Dwarf Mongoose Optimization (DMO) algorithm is a newly proposed intelligent swarm optimization algorithm in recent years. It was developed for continuous optimization problem solutions in its original paper. But real-world problems are not always problems that take continuously variable values. Real-world problems are often problems with discrete variables. Therefore, heuristic algorithms proposed for continuous optimization problems need to be updated to solve discrete optimization problems. In this study, DMO has been updated for binary optimization problems and the Binary DMO (BinDMO) algorithm has been proposed. In binary optimization, the search space consists of binary variable values. Transfer functions are often used in the conversion of continuous variable values to binary variable values. In this study, twelve different transfer functions were used (four Z-shaped, four U-shaped, and four Taper-shaped). Thus, twelve different BinDMO variations were obtained (BinDMO1, BinDMO2, ..., BinDMO12). The achievements of BinDMO variations were tested on thirteen different unimodal and multimodal classical benchmark functions. The effectiveness of population sizes on the effectiveness of BinDMO was also investigated. When the results were examined, it was determined that the most successful BinDMO variation was BinDMO1 (with Z1-shaped transfer function). The most successful BinDMO variation was compared with three different binary heuristic algorithms selected from the literature (SO, PDO, and AFT) on CEC-2017 benchmark functions. According to the average results, BinDMO was the most successful binary heuristic algorithm. This has proven that BinDMO can be chosen as an alternative algorithm for binary optimization problems. [ABSTRACT FROM AUTHOR]

The recently proposed semantic-aligned matching detection transformer (SAM–DETR model) accelerates the convergence of the detection transformer (DETR) by mapping object queries into an identical embedding space as the encoder's output feature map. However, SAM–DETR model has the problem of low detection accuracy compared to other DETR variants. We observe that the lower detection accuracy of SAM–DETR model is caused by the insufficient number of sample points and the inaccurate localization of the sample points during re-sampling, which blurs the generated attention map. This paper proposes an object detector based on a feature correction and double sampling DETR (FCDS-DETR) to solve this problem. FCDS-DETR takes SAM–DETR model as a baseline and builds on it by adding a feature correction module and a double sampling mechanism to achieve further improvement in detection accuracy with a limited number of additional parameters without sacrificing convergence speed. Firstly, FCDS-DETR improves the sampling point localization accuracy by adding a feature correction module to model the inter-channel dependence of the feature maps to be sampled. Secondly, the number of sampled points is increased by the double sampling mechanism, and attention fusion is used to fuse the attention weight maps corresponding to the two sets of sampled points to improve the recognizability of the attention weight maps. The experimental results show that the average precision is improved by +0.7 on the COCO dataset compared with the SAM–DETR model, and the number of parameters is increased by only 10.34 % , which improves the detection performance of the model very well. [ABSTRACT FROM AUTHOR]

Solar photovoltaic energy is the potential energy in the universe for generating electricity and meeting the required load demand. However, on account of partial shading conditions, the difficult task in the PV system is to track global maxima instead of local maxima and maintain the uninterrupted power supply. To solve this problem, a new metaheuristic algorithm is introduced in this paper such as a heap-based optimizer (HBO). The proposed method is developed in MATLAB/Simulink software. The system is examined under distinct irradiation conditions and compared their performance with other methods. The simulation results reveal that the suggested HBO shows a reliable enhancement as compared to other studied methods with regard to tracking maximum power, convergence time, and settling time. The extracted power efficiencies are 99.85% for case 1, 99.96% for case 2, and 99.92% for case 3. It is found that HBO shows better enrichment than other studied methods. [ABSTRACT FROM AUTHOR]

In this paper, we propose a sufficient condition at which a neural network can approximate a set of optimization algorithm solutions; we establish under which conditions a neural network can replace an optimization algorithm to solve a problem with the objective of safely deploying that network in a system where online solutions are necessary to simplify the hardware or allowing the processor to solve the optimization problem on time. To that end, first, we define the family of optimization problems to be addressed; then, we construct a vector with the parameters on which the solution depends, in order to propose a function based on the first-order Karush–Kuhn–Tucker conditions to find conditions under which the inverse of the proposed function maps the problem minimizer with respect to the constructed vector, we provide the sufficiency proof of, both, existence and feasibility of approximation by a neural network regarding the inverse function. Two case studies are proposed, one numerical case showing how a neural network can solve an optimization problem faster than popular solvers to illustrate how it can be implemented in applications where the computation time is tight, and the other case is a Model Predictive Control implementation with the optimization problem solver replaced by a neural network which allows a hardware downgrade; both cases are presented with time statistics comparisons. [ABSTRACT FROM AUTHOR]

Internet of Things (IoT) technology serves many industries to improve their performance. As such, utilizing far distant cloud datacenters to run time-sensitive IoT applications has become a great challenge for the sake of real-time interaction and accurate service delivery time requests. Therefore, the fog computing as a deployment approach of IoT applications has been presented in the edge network. However, inefficient deployment of applications' modules on the fog infrastructure leads to physical resource and bandwidth dissipations, and debilitation of quality of service (QoS), and also increases the power consumption. When all application's modules are highly utilized on a single fog node owing to the reduction in the power consumption, the level of service reliability is decreased. To obviate the problem, this paper takes the concept of fault tolerance threshold into account as a criterion to guarantee applications' running reliability. This paper formulates deployment of IoT applications' modules on fog infrastructure as a multi-objective optimization problem with minimizing both bandwidth wastage and power consumption approach. To solve this combinatorial problem, a multi-objective optimization genetic algorithm (MOGA) is proposed which considers physical resource utilization and bandwidth wastage rate in their objective functions along with reliability and application's QoS in their constraints. To validate the proposed method, extensive scenarios have been conducted. The result of simulations proves that the proposed MOGA model has 18, 38, 9, and 43 percent of improvement against MODCS, MOGWO-I, MOGWO-II, and MOPSO in terms of total power consumption (TPC) and it has 6.4, 15.99, 28.15, and 15.43 dominance percent against them in term of link wastage rate (LWR), respectively. [ABSTRACT FROM AUTHOR]

In this paper, fixed-time convergent RNNs with logarithmic settling time are proposed for solving time-variant quadratic programming (QP), where the domain of attraction of the RNN models can be reasonably estimated a priori. Two novel activation functions are designed, and the exact expressions of the settling time functions for each given initial condition are given; and upper bounds of the settling times for any given initial condition can be obtained. In addition, the generic activation functions are constructed and analyzed to achieve logarithmic fixed-time settling. The RNNs based on the generic activation function are designed, which have a faster convergence rate and a fixed-time convergence property. By estimating the bound of the settling time function under given initial conditions in the bounded region, the semi-global fixed-time convergence of the RNN model is in turn established. It is shown that the predefined-time convergence of modified RNN models can be assured by adopting the inverse of the bound. Simulation results verify effectiveness of the proposed computing schemes for the time-variant QP problem solving and the repetitive motion planning of a redundant manipulator. [ABSTRACT FROM AUTHOR]

The traditional behavior recognition model has the disadvantage that it can't get the internal relationship between similar behaviors, such as smoking, pen, chin and the clamped objects, which limits the actual landing of such fine and complex behaviors as smoking recognition. To solve these problems, this paper puts forward the heterogeneous algorithm HMMA-NET (Heterogeneous multi-task smoking behavior recognition model combined with Attention), which consists of two modules: behavior prior and local detection, aiming at establishing the relationship between behavior and behavior objects. CNN combined with channel attention mechanism is used in both behavior prior module and local detection module. The former uses sign language semantic features to complete the primary prior of behavior according to the obtained behavior affinity vector field, while the latter designs network optimization such as fast Edgebox to obtain candidate areas, so as to transfer component information and achieve the goal of fast fine-grained detection. Finally, the two modules use SaaS mode to complete association recognition. Experiment shows that the algorithm can recognize complex actions effectively, and its accuracy is still equal to or even better than that of a single model, in which the accuracy of detecting smoking behavior scenes is 96.10%, and the false detection rate is 3.6%. The algorithm has been commercialized and applied to the actual monitoring of petrochemical scenes. The running results show that the algorithm can maintain good real-time performance and generalization ability. [ABSTRACT FROM AUTHOR]

With the continuous expansion of the network scale, network technology is also constantly developing. However, with the continuous deterioration of the security environment, the problem of network safety is improving. The traditional single security method has greatly improved the network's stability, but due to the lack of effective cooperation, it becomes increasingly difficult to understand the state changes of the entire network at all times. In such a large environment, research on network security situational awareness can obtain theoretical value and has certain application prospects. The current understanding of cybersecurity situational awareness is not deep enough. Most cases are built in a single-source environment and cannot accurately reflect the perception of attack phases and sequences. To solve this problem, a new model of network safety situation awareness based on multisource data fusion was proposed. The model can effectively perceive the attack stages and sequences and provide an early warning, which is of great importance to improve the network security situation awareness and maintain the network security environment. On the basis of extracting the degree of dissimilarity, in this paper, the fusion-based method is used to generate the attack trajectory, thus forming the multisource data fusion and reconstruction algorithm and finally forming the network security situational awareness model. Compared with the single-source data fusion and reconstruction algorithm, this method has better performance. The final result shows that when the original number of alarms was 1237, after multisource data fusion, the number of alarms was reduced to 124. Moreover, on the basis of multisource data fusion, the detection rate of the number of alarms reached 86.67%, which was 26.67% higher than that of single-source data fusion; the false alarm rate was 5.63%, which was 1.19% lower than that of single-source data fusion. In addition, when using the trajectory reconstruction method to reconstruct the trajectory, the accuracy of the multisource data fusion algorithm was also 1.18% higher than that of the single source, and the completeness also increased by 2.53% compared with the single source. Therefore, the proposed algorithm has higher efficiency, and it is helpful to establish and study the network safety situation consciousness model. [ABSTRACT FROM AUTHOR]

Deformable medical image registration (DMIR) is a crucial task in medical image analysis. Current learning-based approaches concatenate two images into two channels, then use convolution or transformer-based networks to extract features for modeling spatial correspondence. However, this approach mixes the features of the two images before searching for correspondence, making the spatial modeling process challenging. To solve this problem, this paper proposes a purely convolutional framework, DIEGraph, for DMIR. The proposed method first extracts features separately from the moving and fixed images using a dual-branch network. Then, it splits the features into patches and constructs a graph by connecting the nearest neighbors of these patches. Based on the graph representation, the proposed method uses an information exchange graph convolutional network module to exchange dual-branch information and model correspondence. Additionally, a dilated residual fusion module is deployed for feature fusion. We conducted a qualitative and quantitative evaluation on two 3D datasets. The proposed method achieved higher Dice scores than Transmorph by 0.9 % and 3.8 % on the two datasets, respectively, while maintaining a comparable voxel folding percentage. Ablation studies also verify the effectiveness of the components in the proposed model. [ABSTRACT FROM AUTHOR]

Differential evolutionary (DE) algorithm is one of the most frequently used evolutionary computation method for the solution of non-differentiable, complex and discontinuous real value numerical problems. The analytical structure of the mutation and crossover operators used by DE and the initial values of the parameters of the relevant operators affect the problem-solving ability of DE. Unfortunately, there is no analytical method for selecting and initializing the best artificial genetic operators that DE can use to solve a problem. Therefore, there is a need to develop new evolutionary search methods that are parameter-free and insensitive to the artificial genetic operators they use. In this paper, the Bernstein–Levy differential evolution (BDE) algorithm, which has a unique elitist-mutation operator and a Bernstein polynomials-based stochastic parameter-free crossover operator, is introduced. The numerical problem-solving success of BDE is statistically evaluated by using 30 benchmark problems of CEC2014 in the numerical experiments presented. BDE's success in solving the related benchmark problems is statistically compared with six state-of-the-art comparison algorithms. In this paper, three real-world optimization problems are also solved by using the proposed algorithm, BDE. According to statistics generated from the experimental results, BDE is statistically better than comparison methods in solving the related real-world problems. [ABSTRACT FROM AUTHOR]

The main objective of this paper is to present an improved neural network algorithm (INNA) for solving the reliability-redundancy allocation problem (RRAP) with nonlinear resource constraints. In this RRAP, both the component reliability and the redundancy allocation are to be considered simultaneously. Neural network algorithm (NNA) is one of the newest and efficient swarm optimization algorithms having a strong global search ability that is very adequate in solving different kinds of complex optimization problems. Despite its efficiency, NNA experiences poor exploitation, which causes slow convergence and also restricts its practical application of solving optimization problems. Considering this deficiency and to obtain a better balance between exploration and exploitation, searching procedure for NNA is reconstructed by implementing a new logarithmic spiral search operator and the searching strategy of the learner phase of teaching–learning-based optimization (TLBO) and an improved NNA has been developed in this paper. To demonstrate the performance of INNA, it is evaluated against seven well-known reliability optimization problems and finally compared with other existing meta-heuristics algorithms. Additionally, the INNA results are statistically investigated with the Wilcoxon sign-rank test and Multiple comparison test to show the significance of the results. Experimental results reveal that the proposed algorithm is highly competitive and performs better than previously developed algorithms in the literature. [ABSTRACT FROM AUTHOR]

Deep learning techniques can form generalized models that can solve any problem that is not solvable by traditional approaches. It explains the omnipresence of deep learning models across all domains. However, a lot of time is spent on finding the optimal hyperparameters to help the model generalize and give the highest accuracy. This paper investigates a proposed model incorporating hybrid layers and a novel approach for weight initialization aimed at—(1) Reducing the overall trial and error time spent in finding the optimal number of layers by providing the necessary insights. (2) Reducing the randomness in weight initialization with the help of a novel incremental backpropagation based model architecture. The model, along with the principal component analysis-based initialization, substantially provides a stable weight initialization, thereby improving the train and test performance and speeding up the process of convergence to an optimal solution. Furthermore, three data sets were tested on the proposed approach, and they outperformed the state-of-the-art initialization methods. [ABSTRACT FROM AUTHOR]

Prostate diseases often occur in men. For further clinical treatment and diagnosis, we need to do accurate segmentation on prostate. There are already many methods that concentrate on solving the problem of automatic prostate MR image segmentation. However, the design of some hyperparameters of these methods is migrated from the models that are used for nature images which do not consider the difference between medical image and nature image. Besides, there is trend that researchers are likely to use deeper and more complicated networks to achieve high accuracy. The improvement is limited with surging parameters, computations, training time, and inference time. In this paper, we propose an efficient attention residual U-Net to segment the prostate MR image. We analyze the property of prostate MR image and fine-tune the architecture of U-Net. To accelerate the convergence of our method, residual connection and channel attention are added to our network. A set of experiments suggest our method can achieve a similar accuracy of state of the art with less parameters, less computations, shorter training time, and shorter inference time. [ABSTRACT FROM AUTHOR]

In the previous image segmentation methods based on affinity graph learning, it is difficult to obtain clear contour boundaries in the process of image preprocessing, over-segmentation leads to the separation of local regions and most traditional affinity graph learning methods cluster and segment with a single feature of natural images, ignoring the associative characteristics of multiple types of features and cannot effectively utilize the useful information of multiple features of natural images. Aiming at such problems, this paper proposes a tensor-based multi-feature affinity graph learning method for natural image segmentation (TRMFAL). First, the adaptive morphological rebuilding watershed transform is applied to original natural image, the obtained superpixel image contains a lot of boundary contour information, and the fusion of local regions is relatively good; secondly, extract multi-class features from the superpixel blocks in the superpixel image, effectively utilize the features of different characteristics, and integrate them into a multi-feature data matrix according to the corresponding rules; Then, a tensor-based multi-feature affinity graph learning algorithm is proposed, in which tensor is introduced in the algorithm to effectively obtain the higher-order information of image data, and use the projection matrix to embed the original data in the low-dimensional space to decrease the dimension, while minimizing the residual error of each view feature and assign appropriate weights according to the importance of the feature information; finally, use spectral clustering performs clustering and segmentation on affinity graphs to obtain final clustering results and segmented images. In addition, an optimization iterative algorithm based on the alternating multiplier direction method is designed, which effectively solves the problem of solving the TRMFAL model. Sufficient experimental comparisons are carried out on multiple public datasets, and the results prove that the proposed method achieves the optimal clustering performance and segmentation effect. [ABSTRACT FROM AUTHOR]

Few-shot learning studies the problem of classifying unseen images by learning only a small number of samples in these categories with the assistance of a large amount of data in other classes. In recent studies, the idea of transfer learning is an effective method to solve the problem of few-shot classification. However, the insufficient generalization ability of the model still restricts the performance of these transfer-based methods. This paper proposes a masked manifold mixup and fuzzy memory contrastive learning (M3FM) method for transfer-based few-shot learning to improve the generalization ability. We design a regularization technique that enhances the model's learning of local features by masking and mixing the data manifold in the hidden states of neural networks. Then, a momentum updated fuzzy memory is adopted in contrastive learning with the masked mixup manifold to help the model learn the specific distinctions of different categories. Experimental results show that the proposed method outperforms previous baseline methods on miniImageNet, CUB-200, and CIFAR-FS benchmarks. Further adaptation research demonstrates that our method can be generalized to complex few-shot classification tasks and cross-domain scenarios. Ablation studies verify the effectiveness of masked manifold mixup and fuzzy memory contrastive learning. [ABSTRACT FROM AUTHOR]

This paper addresses the reconstruction issue of sparse signal by developing centralized and distributed neurodynamic approaches. An l 1 -minimization problem can be employed for reconstructing sparse signal, and for solving this problem, a centralized neurodynamic approach with derivative feedback is designed. Considering the fact that the distributed approaches decompose the large-scale problem into small-scale one without central processing node in centralized ones, it effectively reduces the computational complexity of a single node. According to the distributed consensus and graph theory, the original l 1 -minimization problem can be equivalently transformed as a distributed optimization problem. Then, based on our proposed centralized approach, a distributed neurodynamic approach with derivative feedback is further proposed. Through the convex optimization theory and Lyapunov method, we indicate that the optimal solution of l 1 -minimization problem is equivalent to the equilibrium point of centralized or distributed approach, and that each neurodynamic approach globally converges to its equilibrium point. Furthermore, by comparing with several state-of-the-art neurodynamic approaches, our proposed approaches demonstrate their effectiveness and superiority by simulation results in reconstructing sparse signals and images. [ABSTRACT FROM AUTHOR]

It is challenging to use unsupervised machine translation models to generate ancient poems. The current method has solved the problems of Under-translation and Over-translation caused by the huge length difference between the translated sentence pairs. However, the above method lacks guidance in generating intermediate vectors, and the denoising ability of the model is very poor. In this paper, we guide vector space distribution during training to improve the quality of the generated ancient poems and the convergence speed of the model. We also introduce the target language information while adding noise, which effectively avoids the recurrence of the Under-translation problem while improving the model's denoising ability. Experiment results on the VP dataset show that our model obtains state-of-the-art results with faster convergence speed. In addition to the BLEU scores, we also made a comparative analysis of ancient poetry sentences generated by different models. The analysis results show that the optimization method proposed in this paper is indeed helpful for generating high-quality ancient poems. [ABSTRACT FROM AUTHOR]

Constructing a personalized learning path is a critical way to solve the problem of cognitive difference and learning disorientation effectively. The construction process of the learning path is closely related to the internal relationship between knowledge and needs to meet different learning scenarios and learning needs. Because of the above requirements, a personalized learning path model based on a knowledge network is proposed in this paper. The algorithm begins by building a knowledge network with learning resource nodes and knowledge points. Following that, the order of the knowledge points was determined using their sequential link. A sequence of learning materials that adhere to user characteristics was eventually acquired by evaluating the learning time limit of various learning contexts. The proposed approach was tested on the data sets of open MOOPer and Python learning platforms. Compared with traditional learning path construction algorithms, the proposed algorithm improves the accuracy and relevance. [ABSTRACT FROM AUTHOR]

Traffic sign detection is a challenging task for the unmanned driving system, especially for the detection of multi-scale targets and the real-time problem of detection. In the traffic sign detection process, the scale of the targets changes greatly, which will have a certain impact on the detection accuracy. Feature pyramid is widely used to solve this problem, but due to the diversity of traffic sign sizes, it cannot accurately extract multi-size feature maps, thus destroying the feature consistency between traffic signs. Moreover, in practical application, it is difficult for common methods to improve the detection accuracy of multi-scale traffic signs while ensuring real-time detection. In this paper, we propose an improved feature pyramid model, named AF-FPN, which utilizes the adaptive attention module (AAM) and feature enhancement module (FEM) to reduce the information loss in the process of feature map generation and enhance the representation ability of the feature pyramid. We replaced the original feature pyramid network in YOLOv5 with AF-FPN, which improves the detection performance for multi-scale targets of the YOLOv5 network under the premise of ensuring real-time detection. Furthermore, a new automatic learning data augmentation method is proposed to enrich the dataset and improve the robustness of the model to make it more suitable for practical scenarios. Extensive experimental results on the Tsinghua-Tencent 100 K (TT100K) dataset demonstrate that compared with several state-of-the-art methods, our method is more universal and superior. [ABSTRACT FROM AUTHOR]

Approximate nearest neighbor (ANN) search in high-dimensional spaces is fundamental in many applications. Locality-sensitive hashing (LSH) is a well-known methodology to solve the ANN problem. Existing LSH-based ANN solutions typically employ a large number of individual indexes optimized for searching efficiency. Updating such indexes might be impractical when processing high-dimensional streaming data. In this paper, we present a novel disk-based LSH index that offers efficient support for both searches and updates. The contributions of our work are threefold. First, we use the write-friendly LSM-trees to store the LSH projections to facilitate efficient updates. Second, we develop a novel estimation scheme to estimate the number of required LSH functions, with which the disk storage and access costs are effectively reduced. Third, we exploit both the collision number and the projection distance to improve the efficiency of candidate selection, improving the search performance with theoretical guarantees on the result quality. Experiments on four real-world datasets show that our proposal outperforms the state-of-the-art schemes. [ABSTRACT FROM AUTHOR]

Buried pipeline is threatened from the soil displacement due to earthquake and other causes, which leads to the formation of unexpected external forces such as bending moment. The problem can be worse with the appearance of defects, resulting in the reduction in pipe capacity. The paper focuses on the overall problem of defected pipe crossing the strike-slip fault. A full-scaled FE model can be very complicated with the large-scale and micro-scale levels corresponding to the strike-slip fault and defect on pipe problems, respectively. To ease this difficulty, the macro and micro problems are solved separately with two types of FE models and their corresponding databases. To be specific, one FE model is used for predicting external moment due to strike-slip fault and the other is for predicting the moment capacity of defected pipe. Data-driven models are consequently developed with artificial neural network (ANN) for each database generated from these types of models: ANN1 evaluating moment capacity of defected pipe (R2 is 0.9943 on test set) and ANN2 predicting both moment and axial force appeared in pipe due to strike-slip fault (R-squares are 0.9883 and 0.9929 on test set, respectively). Consequently, the stress–strength analysis for the overall problem is solved. Accounting for the unavoidable uncertainty of the models, the paper proposed an approach which assumes that the actual distribution of residual of a model is equivalent to this of the test set. The distributions of residuals on test set of these ANNs are tested to be normally distributed and generated by the conventional Monte Carlo simulation. To the end, the deterministic problem leads to the failure probability. The proposed framework has been investigated, and the final results on this selective parametric study are reasonable. [ABSTRACT FROM AUTHOR]

The problem of discrete-time dynamic constrained quadratic programming including equality and inequality constraints is formulated and investigated, simply termed future constrained quadratic programming (FCQP) problem in this paper. To obtain the optimal solution of such an FCQP problem in real time and with high precision, a novel discrete-time zeroing neurodynamic (DTZN) algorithm, which is developed by combining the corresponding continuous-time zeroing neurodynamic model and a five-step explicit linear multi-step (ELMS) rule, is proposed and termed ELMS-type five-step DTZN (5SDTZN) algorithm. Then, the convergence and precision of the ELMS-type 5SDTZN algorithm are analyzed theoretically. For comparison, three other DTZN algorithms as well as discrete-time gradient and varying-parameter neurodynamic algorithms are also presented. Afterward, through numerical verifications and comparisons, the efficacy and superiority of the ELMS-type 5SDTZN algorithm for solving the FCQP problem are illustrated. Finally, the applicability of the ELMS-type 5SDTZN algorithm for the coordinated repetitive motion control of a wheeled mobile robot with physical constraints is demonstrated. [ABSTRACT FROM AUTHOR]

This paper proposes an improved version of the artificial ecosystem-based optimization (AEO) algorithm called artificial ecosystem-based optimization–opposition-based learning (AEO-OBL), with the aim of improving the performance of the original AEO. In addition, it is utilized for determining the optimal allocation of distributed generation (DG) units in radial distribution networks (RDNs) with the aim of minimizing power and energy losses. The stochastic nature of renewable DGs such as wind turbine and photovoltaic generation is taken in consideration using appropriate probability models. The Loss Sensitivity Index is used to assess the most suitable busses for the integration of DG units in the RDN. AEO is nature-inspired optimization algorithm which imitates the flow of energy in an ecosystem on earth. In the proposed AEO-OBL, the search ability and the balance between the exploration and exploitation phases in the original AEO are enhanced. In the AEO-OBL, five efficient strategies are used to avoid falling on a local optimal: (1) enhanced linear weight coefficient a, (2) production operator, (3) modified consumption operator, (4) modified decomposing operator and (5) opposition-based learning (OBL). The performance of the proposed technique is validated on IEEE 33-bus and 85-bus RDNs. To emphasize the superiority of the proposed technique, the results are compared with the original AEO, Henry gas solubility optimizer (HGSO) and Harris hawks optimization (HHO) algorithm results. Besides, the developed algorithm is compared with other optimization algorithms in literature that solved the same problem. The outcomes indicate a better performance of AEO-OBL relative to other algorithms. Accordingly, AEO-OBL can be a very suitable algorithm in solving the problem of optimal DG allocation in RDNs. [ABSTRACT FROM AUTHOR]

Evolutionary multi-objective optimization (EMO) refers to the domain in which an evolutionary algorithm is applied to tackle an optimization problem with multiple objective functions. The literature is rich with many approaches proposed to solve multi-objective problems including the NSGA-II, MOEA/D, and MOPSO algorithms. The proposed approaches include stand-alone as well as hybrid techniques. One critical aspect of any evolutionary algorithm (EA) is the stopping criterion. The selection of a specific stopping criterion can have a considerable effect on the performance and the final solution provided by the EA. A number of different stopping criteria, specifically designed for EMO, have been proposed in the literature. In this paper, the performance of six different EMO algorithms is tested and compared using four stopping criteria. The experiments are performed using the ZDT, DTLZ, CEC2009, Tanaka and Srivana test functions. Experimental results are analyzed to highlight the proper stopping criteria for different algorithms. [ABSTRACT FROM AUTHOR]

Graph Neural Networks (GNNs) have shown excellent performance in graph-related tasks and have arisen widespread attention. However, most existing works on GNNs mainly focus on proposing a novel GNN model or modifying the models to improve performance in various graph-related tasks and seldom consider possible problems in the graph data as model input data. This paper studies that low-degree nodes, accounting for most of the real-world graphs, naturally have message passing insufficiency problem due to too little information from other nodes when generating node embedding, which will affect the performance of the GNNs. To solve this problem, we propose a simple but practical method-Optimize Graph Then Training (OGT), which adds edges between low-degree nodes and some nodes with the same predicted label based on the GNNs prediction results and the inherent information in the graph. The OGT aims to improve the performance of GNNs in semi-supervised node classification tasks by augmenting the input data. More importantly, the OGT is regarded as a data preprocessing technique and can be used naturally with baseline GNN models (e.g., GCN, GAT, GraphSAGE, and SGC) to improve the performance of these models without making other modifications. Extensive experiments on three benchmark citation datasets with five typical GNN models verify that OGT consistently improves the performance of various GNNs to a great extent, where 1.9% (Cora), 1.0% (Citeseer), and 1.3% (Pubmed) average accuracy improvement on the node classification task. [ABSTRACT FROM AUTHOR]

Selecting relevant feature subsets is vital in machine learning, and multiclass feature selection is harder to perform since most classifications are binary. The feature selection problem aims at reducing the feature set dimension while maintaining the performance model accuracy. Datasets can be classified using various methods. Nevertheless, metaheuristic algorithms attract substantial attention to solving different problems in optimization. For this reason, this paper presents a systematic survey of literature for solving multiclass feature selection problems utilizing metaheuristic algorithms that can assist classifiers selects optima or near optima features faster and more accurately. Metaheuristic algorithms have also been presented in four primary behavior-based categories, i.e., evolutionary-based, swarm-intelligence-based, physics-based, and human-based, even though some literature works presented more categorization. Further, lists of metaheuristic algorithms were introduced in the categories mentioned. In finding the solution to issues related to multiclass feature selection, only articles on metaheuristic algorithms used for multiclass feature selection problems from the year 2000 to 2022 were reviewed about their different categories and detailed descriptions. We considered some application areas for some of the metaheuristic algorithms applied for multiclass feature selection with their variations. Popular multiclass classifiers for feature selection were also examined. Moreover, we also presented the challenges of metaheuristic algorithms for feature selection, and we identified gaps for further research studies. [ABSTRACT FROM AUTHOR]

The paper aims to introduce the novel concept of q-connection number (q-CN) for interval-valued q-rung orthopair fuzzy set (IVq-ROFSs) and thus to develop a method for solving the multiple-attribute group decision making (MAGDM) problem. The IVq-ROFS is a tool to represent the uncertain information with an integer parameter q ≥ 1 , while the connection number (CN) processes the uncertainties and certainties into a single system with three degrees, namely "identity", "contrary" and "discrepancy". Driven by these required properties, this paper introduces a q-CN for IVq-ROFSs to represent the information in a more concise way. To this end, we divide the paper into three aspects. First, we define q-CN and a scoring function to evaluate the numbers. Second, we give some new q-exponential operation laws (q-EOLs) and operators over q-CNs in which bases are real numbers and exponents are q-CNs. Moreover, we define an operator based on these laws and derive their properties. Third, a novel MAGDM method for solving decision problems with IVq-ROFS information is illustrated with several examples. The advantages and superiority analysis of the proposed framework are also given to assert the results. [ABSTRACT FROM AUTHOR]

This paper addresses a new multi-level serial closed-loop supply chain (CLSC) model with an integrated formulation of batch deliveries, quality-dependent return rate, random defective rate, rework process, and learning effects. The vendor managed inventory contract has motivated companies to cope with the bullwhip effect and its associated shortages. However, getting precise information on manufacturing/remanufacturing time is a challenge owing to human learning. Accordingly, this paper tries to incorporate the learning effect on the inventory control problem. This is a bridge to fill the gaps in the literature by considering how the learning process affects the production time and system costs in a CLSC problem. Moreover, the mean-standard deviation utility function is used in order to take the random defective rate into account, to minimize mean system costs with respect to the standard deviation. Due to the complexity of the proposed mixed-integer nonlinear programming model, three metaheuristic algorithms named, genetic algorithm, invasive weed optimization algorithm, and moth flame optimization algorithm are used to solve the problem, and Fibonacci algorithm will examine the validity of the applied methods. Two sets of numerical examples in small, medium, and large sizes are presented to illustrate the performance of the algorithms in terms of objective function values and required CPU-time. Afterward, a statistical analysis is utilized in order to evaluate the proposed algorithms. Finally, the results show how significantly the learning effect in manufacturing/remanufacturing time affects the costs of CLSC problem. [ABSTRACT FROM AUTHOR]

It has been witnessed that there is an increasing interest in video salient object detection (VSOD) in computer vision field. Different from image salient object detection (ISOD), VSOD not only requires appearance information but also needs motion cues. Thus, it is essential to exploit spatiotemporal information to generate accurate saliency results. Existing VSOD models mainly combine an ISOD model with long short-term memory (LSTM) or flow-estimation modules to integrate saliency cues estimated from spatial and temporal domain. However, flow-estimation modules heavily rely on optical flow images; the generation process of which is rather time-consuming and severely limits its applications in practice. Besides, the LSTM can only exploit motion cues via a step-by-step propagation in the time domain and is hard to realize the multi-scale spatiotemporal interaction. In this paper, we propose the SCANet to solve the above problems. Specifically, we develop the pyramid dilated 3D convolutional (PD3C) module to generate rich temporal features by leveraging context information. Besides, a feature aggregation module is designed to effectively integrate spatial and temporal features. Equipped with these modules, the SCANet is capable of generating high-quality saliency maps at more than real-time inference speed (41 FPS on a single Titan Xp GPU). Extensive experimental results on six widely used benchmark datasets prove that SCANet outperforms state-of-the-art methods in terms of three standard evaluation metrics. Our code will be publicly available at https://github.com/clelouch/SCANet. [ABSTRACT FROM AUTHOR]

A lot of research has validated that zeroing neural network (ZNN) model is a reliable tool for solving time-varying problems. Generally, convergent performance is often one of the most important measures for the practicability of the ZNN model. Besides, its practicability is also up to its anti-noise ability in the process of solving time-varying problems. It is worth noting that numerous existing ZNN models perform well in terms of the above two issues. In order to further enhance the robustness and convergence of the ZNN model, two novel activation functions (novel activation function-1 (NAF-1) and novel activation function-2 (NAF-2)) are designed. Based on the above two NAFs, two robust ZNN models (zeroing neural network-1 (ZNN-1) and zeroing neural network-2 (ZNN-2)) with fixed time convergence and strong robustness to noise for solving time-varying Sylvester equation (TVSE) are proposed in this paper. The robustness and convergence of the two proposed ZNN models are verified by rigorous mathematical analysis, and their practicability are validated by three simulation experiments (second-order and third-order TVSE solving and manipulator trajectory tracking). All the studies indicate that the two ZNN models can quickly find the accurate solution of TVSE in noiseless and noisy environment. [ABSTRACT FROM AUTHOR]

In many real-world applications, collecting and labeling the data is expensive and time-consuming. Thus, there is a need to obtain a high-performance learner by leveraging the data or knowledge from other domains. Transfer learning is a promising method to solve the above problems. In this paper, we propose a multitask transfer learning method, which aims to improve the performance of the target learner by transferring knowledge from the related source tasks. First, we formulate the target learner as a nonlinear function, which is approximated by the linear combination of the eigenfunctions. Further, to transfer knowledge from the source tasks, we constrain the target model to be the linear combination of the source models according to the previous work. However, knowledge from some source tasks may not be useful for adaptation, so we add a sparse constraint to the objective function to select the related source tasks. Different from previous transfer learning methods, our method transfers knowledge by jointly learning the source tasks and the target task. Besides, it can select the source tasks associated with the target task by the sparse constraint. Empirically, the method exhibits protection against negative transfer. Finally, we compare our proposed method with three single-task learning methods and six state-of-the-art multitask learning methods on two data sets. When compared with the second best results, the nMSE of our method achieves a relative improvement of 10.85 % with a training size of 100 on the SARCOS data set and a relative improvement of 4.26 % with a training ratio of 20 % on the Isolet data set. Experimental results show that our proposed method can effectively improve the performance of the target task by transferring knowledge from the related source tasks. [ABSTRACT FROM AUTHOR]

Conventional data simplification algorithms depend much on scanning technology. However, with the development of the scanning technology, the conventional algorithm is unable to process numerous redundant data, leading to increased noise of point cloud data, failure of locating split center, and low simplification efficiency. In order to remove the noise in point cloud data, improve randomness for choosing the segmentation center, and obtain robust curvature information of the point cloud data and accurate segmentation results, the following improvements are made to the conventional spatial segmentation algorithms in this paper. Firstly, on the basis of denoising by B-spline wavelet filtering, voxel grid of point cloud data is carried out to select the effective segmentation center. Secondly, through the improvement of the subspace clustering algorithm, reasonable space division can be realized. Thirdly, based on the first-order neighbourhood surface of the vertices of the mesh surface, the curvature of sample points can be estimated. Fourthly, by introducing the distance between the neighbourhood sample and the target sample, the distance between the two points is taken as the geodesic distance, the smooth correction of the estimated result can be realized and the finally simplified point cloud segmentation result can be obtained. Finally, a robust spatial data simplification method is implemented. The experimental results show that this method can further segment spatial data and obtain simplified segmentation results while ensuring segmentation efficiency. Compared with the unified sampling method, the curvature simplification method, the isometric sampling method, and the random sampling method, the proposed method can reduce the sensitivity to boundary noise data, solve the problem of fuzzy boundary division, and overcome the disadvantage of poor accuracy caused by too high dimension. [ABSTRACT FROM AUTHOR]

The purpose of smart city is to enhance the optimal utilization of scarce resources and improve the resident's quality of live. The smart cities employed Internet of Things (IoT) to create a sustainable urban life. The IoT devices such as sensors, actuators, and smartphones in the smart cities generate data. The data generated from the smart cities are subjected to analytics to gain insight and discover new knowledge for improving the efficiency and effectiveness of the smart cities. Recently, the application of deep learning in smart cities has gained a tremendous attention from the research community. However, despite raise in popularity and achievements made by deep learning in solving problems in smart cities, no survey has been dedicated mainly on the application of deep learning in smart cities to show recent progress and direction for future development. To bridge this gap, this paper proposes to conduct a dedicated survey on the applications of deep learning in smart cities. In this paper, recent progress, new taxonomies, challenges and opportunities for future research opportunities on the application of deep learning in smart cities have been unveiled. The paper can provide opportunities for experts in the research community to propose a novel approach for developing the research area. On the other hand, new researchers interested in the research area can use the paper as an entry point. [ABSTRACT FROM AUTHOR]

Meta-heuristic algorithms have been proposed to solve several optimization problems in different research areas due to their unique attractive features. Traditionally, heuristic approaches are designed separately for discrete and continuous problems. This paper leverages the meta-heuristic algorithm for solving NP-hard problems in both continuous and discrete optimization fields, such as nonlinear and multi-level programming problems through extensive simulations of volcano eruption process. In particular, a new optimization solution named volcano eruption algorithm is proposed in this paper, which is inspired from the nature of volcano eruption. The feasibility and efficiency of the algorithm are evaluated using numerical results obtained through several test problems reported in the state-of-the-art literature. Based on the solutions and number of required iterations, we observed that the proposed meta-heuristic algorithm performs remarkably well to solve NP-hard problem. Furthermore, the proposed algorithm is applied to solve some large-size benchmarking LP and Internet of vehicles problems efficiently. [ABSTRACT FROM AUTHOR]

Facial expression recognition (FER) in the wild is very challenging due to occlusion, posture, illumination, and other uncontrolled factors. Learning discriminant features for FER using Convolutional Neural Networks is a momentous task for the significant class imbalance, wrong labels, inter-class similarities, and intra-class variations. The traditional method utilizes the Cross entropy loss function to optimize the convolutional network to obtain discriminative features for classification. However, this loss function cannot effectively solve the above problems in practice and cannot contribute to obtaining highly discriminant facial features for further analysis. Center loss improves the learning efficiency by reducing the intra-class distance of similar expressions, while the improvement of inter-class similarity, class imbalance, and generalization is insufficient. In this paper, we propose a lightweight Effective Attention Feature Reconstruction loss (EAFR loss), which can further optimize the feature space and enhance the discriminability of expression. The loss model is composed of the Focal Smoothing loss (FS loss) and the Aggregation-Separation loss (AS loss). Firstly, the FS loss can improve the poor recognition performance caused by imbalanced classes and prevent paranoid knowledge learning behaviors. Meanwhile, AS loss further accurately condenses the intra-class expression features and expands the inter-class distance, which is achieved by using progressive stage max-pooling channel and position attention mechanism and lightweight asymmetric autoencoder model for feature reconstruction. Finally, the EAFR loss joins the above two loss functions to more comprehensively solve the above typical problems for FER in the wild. We validate the proposed loss function on three most commonly used large-scale wild expression datasets (RAF-DB, FERPlus, and AffectNet), and the results show that our model achieves superior performance to several state-of-the-art methods. [ABSTRACT FROM AUTHOR]

The goal of session-based recommendation is to make item predictions at the next timestamp based on the anonymous ongoing session. Previous work mainly models user's preference by exploring the transition pattern between the interacted items in the session. However, they generally fail to pay enough attention to the item importance in terms of the relevance of the items to user's main purpose. This paper proposes a Session-based Recommendation approach with an Importance Extraction Module, i.e., SR-IEM improved , which can simultaneously consider user's long-term interactions and recent behavior in current session. Specifically, we modify the self-attention mechanism to avoid introducing bias for estimating the importance of each item in the session. Then, the item importances are utilized to produce user's long-term preference, and the sequential signals are incorporated in the long-term interest modeling. Next, the long-term preference and user's current interest which is conveyed by the last interacted item in the session are combined to obtain user's final preference representation. Finally, item predictions are generated using the user preference, where a normalization layer is adopted to solve the long-tail problem. Extensive experiments are conducted on three public benchmark datasets, i.e., Yoochoose 1/64, Yoochoose 1/4 and Diginetica. The experimental results show that SR-IEM improved can outperform the start-of-the-art baselines in terms of Recall and MRR for session-based recommendation. In addition, compared to the state-of-the-art neural methods, SR-IEM improved can obviously reduce the computational complexity. [ABSTRACT FROM AUTHOR]

The Vehicle Routing Problem (VRP) is a class of well-known combinatorial optimization problems. The great interest in the VRP is due to its practical importance, as well as the difficulty in solving it. The Capacitated Vehicle Routing Problem (CVRP) is the most common variant of the VRP. Most of the approaches that have been developed to solve this problem tend to solve the problem in a centralized way. There has been very little research done into solving this problem in a distributed manner. In this paper, we propose an innovative approach for solving CVRP in a distributed manner based on multi-agent systems and using the game theory, which consists of three types of intelligent agents: customer agent, vehicle agent, and depot agent and includes the two phases: cluster construction and clusters optimization. The cluster construction phase consists of playing a Game Theoretic Clustering Algorithm by customer agents to divide the customers into several clusters, each consisting of a cluster head and several member customers. Then, we present a Game Theoretic Clusters Optimization Algorithm for the cluster optimization phase, which is played by vehicle agents to minimize the cost of each vehicle's route to optimize the quality of the solution. The performance of the proposed approach is evaluated on 48 instances from 10 standard benchmark sets and compared with some state-of-the-art methods in terms of execution time and quality of solutions. Our experiments illustrated that the proposed method can compete or even outperform much more complex algorithms. [ABSTRACT FROM AUTHOR]

Recently, various cross-domain recommendation (CDR) models are proposed to overcome the sparsity problem, which leverage relatively abundant rating data from the auxiliary domain to improve recommendation performance of target domain. Though matrix factorization-based collaborative filtering algorithms gain dominance in single-domain recommendation systems, they cannot be used directly in cross-domain cases as the obtained latent factors of the target and auxiliary domains may not be aligned, which will lead to inaccurate knowledge transfer from the auxiliary domain to the target one. A CDR model named CDCFLFA is presented in this paper to solve this problem. In CDCFLFA, firstly latent factors between the two domains are aligned based on pattern matching. Then, user preferences of the auxiliary domain are transferred to update the original user latent vectors of target domain. Finally, a linear least square problem is solved to compute the item latent vectors of target domain and thus unknown ratings are obtained according to the updated user and item latent vectors. CDCFLFA does not require the same user or item sets between the two domains. Extensive experiments are conducted, and the results show that CDCFLFA achieves smaller MAE and RMSE values and larger precision and recall than the previous single- and cross-domain recommendation methods. Hence, CDCFLFA can be regarded as an effective cross-domain extension of single-domain matrix factorization algorithm. [ABSTRACT FROM AUTHOR]

The existing view-based 3D object classification and recognition methods ignore the inherent hierarchical correlation and distinguishability of views, making it difficult to further improve the classification accuracy. In order to solve this problem, this paper proposes an end-to-end multi-view dual attention network framework for high-precision recognition of 3D objects. On one hand, we obtain three feature layers of query, key, and value through the convolution layer. The spatial attention matrix is generated by the key-value pairs of query and key, and each feature in the value of the original feature space branch is assigned different importance, which clearly captures the prominent detail features in the view, generates the view space shape descriptor, and focuses on the detail part of the view with the feature of category discrimination. On the other hand, a channel attention vector is obtained by compressing the channel information in different views, and the attention weight of each view feature is scaled to find the correlation between the target views and focus on the view with important features in all views. Integrating the two feature descriptors together to generate global shape descriptors of the 3D model, which has a stronger response to the distinguishing features of the object model and can be used for high-precision 3D object recognition. The proposed method achieves an overall accuracy of 96.6% and an average accuracy of 95.5% on the open-source ModelNet40 dataset, compiled by Princeton University when using Resnet50 as the basic CNN model. Compared with the existing deep learning methods, the experimental results demonstrate that the proposed method achieves state-of-the-art performance in the 3D object classification accuracy. [ABSTRACT FROM AUTHOR]

To be in the competitive industrial world, industries required high quality, speed in completing the required work, and commitment to the delivery dates. One of the most important issues in the field of production management is the job shop scheduling problem (JSSP). In this paper, the researchers tried to solve JSSP of factory by presenting a method to improve the factory's production. Job shop scheduling (JSS) is a suitable method for solving these types of problems, which aims to improve the production flow through minimizing the whole operation time of the products. Moreover, considering the factory that depends on workers as same as machine, human factor should be considered while scheduling by using the workers' weightage, in order to improve the workers' working time flexibility in terms of their waiting time among their tasks by proposed model of JSS. In addition, the researchers proposed a new combination of weightage values by using Taguchi method, regarding to improve the workers' working time and using genetic algorithm (GA) to solve the proposed model of JSS. One of the factories which is located in Jordan, and it is considered as one of the important factories; nevertheless, it can cover the local demands hardly, and hence, it deserves to be as a study case for this research. The findings of the studies decreased the whole operation time of the products by saving 75 min for each production line and 90 min by using GA, and the proposed model improved the flexibility of the workers' working time in terms of their waiting times among their tasks. [ABSTRACT FROM AUTHOR]

The most existing sign language recognition methods have made significant progress. However, there are still problems in the field of sign language recognition: Traditional SLR technology relies on external devices such as data gloves, position tracker, and has achieved limited success. Moreover, the current state-of-the-art vision-based technologies cannot be applied in practice due to the difficulty in balancing accuracy and speed, because most of them pay the cost of running time for better sign language classification accuracy. In this paper, we propose a (2+1)D-SLR network based on (2+1)D convolution, which is different from other methods in that the proposed network can achieve higher accuracy with a faster speed. Because (2+1)D-SLR can learn spatio-temporal features from the raw sign RGB frames. In addition, the existing Chinese sign language dataset is difficult to guarantee the personality differences between different sign language speakers and the presentation differences of the same presenter. Therefore, we propose a large-scale Chinese sign language video dataset called NCSL to solve this problem, including 300 different sign language vocabulary which demonstrated by 30 volunteers, 10 times each. We also validated our method on NCSL and another large-scale sign language dataset, i.e., LSA64, Achieved 96.4% and 98.7% accuracy, respectively, demonstrating that our method can not only achieve competitive accuracy but be much faster than current well-known sign language recognition methods. [ABSTRACT FROM AUTHOR]

Named entity recognition (NER) is an important task in natural language processing. NER is usually used to classify documents, extract information, and translate languages. However, few studies have used NER types to automatically generate questions. In this paper, we proposed a method named NER2QUES to solve the above problem for a low-resource language such as Vietnamese. NER2QUES was the combining pre-trained language model and sequence-to-sequence model. Specifically, we used BERT to detect NERs in a sentence and then applied a sequence-to-sequence model to automatically generate questions that corresponded to NER's types. We compared the accuracy of the proposed method to PhoBERT and spaCy on the NER task. Also, we used F1, BLEU, ROUGE, and METEOR to measure the effectiveness of this approach with the rules-based method, T5, and BERT on question generation tasks. The experiment results show that the accuracy of our method is more improved than previous methods' accuracy of 94% on SQuAD, 89% on XQuAD, and 95% on MLQA. This indicates that using NER to automatically generate questions may enrich question answering systems. [ABSTRACT FROM AUTHOR]

Tensor decomposition is widely used in a variety of applications such as data mining, biomedical informatics, neuroscience, and signal processing. In this paper, we propose a deep non-negative tensor factorization (DNTF) model to learn intrinsic and hierarchical structures from multi-way data. The DNTF model takes the Tucker tensor decomposition as a building block to stack up a multi-layer structure. In such a way, we can gradually learn the more abstract structures in a higher layer. The benefit is that it helps to mine intrinsic correlations and hierarchical structures from multi-way data. The non-negative constraints allow for clustering interpretation of the extracted data-dependent components. The objective of DNTF is to minimize the total reconstruction loss resulting from using the core tensor in the highest layer and the mode matrices in each layer to reconstruct the data tensor. Then, a deep decomposition algorithm based on multiplicative update rules is proposed to solve the optimization problem. It first conducts layer-wise tensor factorization and then fine-tunes the weights of all layers to reduce the total reconstruction loss. The experimental results on biosignal sensor data demonstrate the effectiveness and robustness of the proposed approach. [ABSTRACT FROM AUTHOR]

This paper presents an accelerated gradient-based neural network (GNN) to achieve visual servoing of a surgical endoscope robot. A KUKA LWR 4+ robot with seven joints is used to serve as an endoscope holder. Kinematic mapping is established between the joint space of the robot and the image space of the camera. For surgical applications, the motions of the KUKA robot are constrained with respect to a remote-center-of-motion (RCM) point. Meanwhile, each joint of the KUKA robot has its own physical limits (e.g., joint-angle and joint velocity limits) that cannot be violated. By taking into account the kinematic equation, RCM constraints and physical limits, a control scheme possessing a quadratic programming (QP) formulation is constructed. To solve the QP problem, an inverse-free GNN model is accelerated to be finite-time convergent using a powerful activation function. Mathematical derivations of the accelerated GNN model and theoretical proofs relevant to the finite-time convergence are detailed. Comparative validations are conducted with the superior convergence performance of the accelerated GNN model substantiated. The effectiveness of the proposed GNN solution for vision-based control of the surgical endoscope is verified with RCM constraints and physical limits respected simultaneously. [ABSTRACT FROM AUTHOR]

Cloud providers deliver heterogeneous virtual machines to run complicated jobs submitted by users. The task scheduling issue is formulated to a discrete optimization problem which is well-known NP-Hard. This paper presents a hybrid meta-heuristic algorithm based on genetic and thermodynamic simulated annealing algorithms to solve this problem. In the proposed algorithm, the genetic and simulated annealing algorithms have respective global and local search inclinations covering each other's shortcomings. A novel theorem is presented and applied to produce a semi-conducted initial population. In a used genetic algorithm with a global trend, the crossover operator is performed to explore search space. The thermodynamic simulated annealing algorithm is utilized to improve the efficiency, which considers entropy and energy difference concepts in the cooling schedule process. After obtaining a suitable solution, one of the three novel neighbor operators is randomly called to enhance the given solution potentially. In this way, the efficient balance between exploration and exploitation in the search space is achieved. Simulation results prove that the proposed hybrid algorithm has 10.17%, 9.31%, 7.76%, and 8.21% dominance in terms of makespan, schedule length ratio, speedup, and efficiency against other comparative algorithms. [ABSTRACT FROM AUTHOR]

In this paper, an image cipher is presented based on DNA sequence operations, image filtering and memrisitve chaotic system. Firstly, plain image is preprocessed by a self-updating transformation based on dynamic image filtering (STDIF), which may solve the problem that conventional filtering is invalid for special image such as all-zeros. Subsequently, the resulting image is converted into the DNA sequences by the dynamic DNA encoding rules, and the rules are produced by plain image information and chaotic system. Next, the DNA matrix is permutated by a double random 3D matrix scrambling (DR3DMS), thus every element may arbitrarily move to other position. Thereafter, the permutated matrix is diffused by a plane diffusion of 3D DNA matrix controlled by plain image information (PD3DPI) to resist statistical attacks. Finally, the obtained DNA matrix is transformed into cipher image via DNA decoding rules. Additionally, SHA 256 hash function value of plain image is applied to choose keystreams from the obtained chaotic sequences, which not only solves the problem of repeatedly generating chaotic sequences when encrypting different images, but also results into higher difficulty for hackers to decipher the method by use of plaintext attack. Experimental results and security analyses demonstrate its security and effectiveness. [ABSTRACT FROM AUTHOR]

Suspension bridges are critical components of transport infrastructure around the world. Therefore, their operating conditions should be effectively monitored to ensure their safety and reliability. However, the main cables of suspension bridges inevitably deteriorate over time due to corrosion, as a result of their operational and environmental conditions. Thus, accurate annual corrosion rate predictions are crucial for maintaining reliable structures and optimal maintenance operations. However, the corrosion rate is a chaotic and complex phenomenon with highly nonlinear behavior. This paper proposes a novel predictive model for the estimation of the annual corrosion rate in the main cables of suspension bridges. This is a hybrid model based on the multilayer perceptron (MLP) technique optimized using marine predators algorithm (MPA). In addition, well-known metaheuristic approaches such as the genetic algorithm (GA) and particle swarm algorithm (PSO) are employed to optimize the MLP. In order to implement the proposed model, a comprehensive database composed of 309 sample tests on the annual corrosion rate from all around the world, including various factors related to the surrounding environmental properties, is utilized. In addition, several input combinations are proposed for investigating the trigger factors in modeling the annual corrosion rate. The performance of the proposed models is evaluated using various statistical and graphical criteria. The results of this study demonstrate that the proposed hybrid MLP-MPA model provides stable and accurate predictions, while it transcends the previously developed approaches for solving this problem. The effectiveness of the MLP-MPA model shows that it can be used for further studies on the reliability analysis of the main cables of suspension bridges. [ABSTRACT FROM AUTHOR]