Content of Theory, Research and Development in our journal

        Published in last 1 year |  In last 2 years |  In last 3 years |  All
    Please wait a minute...
    For Selected: Toggle Thumbnails
    T-Transformer Model for Predicting Tensor Time Series
    LI Wen, CHEN Jiawei, LIU Ruixue, HOU Yuguo, DU Shouguo
    Computer Engineering and Applications    2023, 59 (11): 57-62.   DOI: 10.3778/j.issn.1002-8331.2211-0168
    Abstract20)      PDF(pc) (570KB)(13)       Save
    Tensor time series shows the co-evolution characteristics of multiple time series with time changes, and tensor time series prediction has become an important problem. Aiming at the high-dimensional characteristics of tensor time series, T-Transformer, a tensor time series prediction method based on Transformer, is proposed, which integrates tensor operation and Transformer into a unified framework. The method first uses tensors to represent higher-order time series, uses tensor slices and vectorization to convert tensor time series into vectors, and then inputs the vectors into Transformer model after coding, and finally obtains the predicted value of tensor time series. Experiments on three open data sets show that the proposed method achieves better prediction results than the benchmark method.
    Reference | Related Articles | Metrics
    Improved Policy Optimization Algorithm Based on Curiosity Mechanism
    ZHANG Qiyang, CHEN Xiliang, CAO Lei, LAI Jun
    Computer Engineering and Applications    2023, 59 (11): 63-70.   DOI: 10.3778/j.issn.1002-8331.2205-0060
    Abstract17)      PDF(pc) (690KB)(9)       Save
    In the generation process of reinforcement learning decision model, due to the complex environment and incomplete observation of state information, the classical proximal policy optimization algorithm faces problems such as low exploration and utilization efficiency and poor effect of generated strategies, this paper proposes an MNAEK-PPO(proximal policy optimization based on maximum number of arrival & expert knowledge algorithm) based on curiosity mechanism. Focusing on the difficult problem of exploring the strategy space, by constructing the exploration frequency matrix of the agent in the training process, the exploration frequency is treated as an internal reward to participate in the agent’s reinforcement learning and training process. In addition, expert knowledge is added to assist the agent in making decisions. Through experiments in the intelligent battlefield simulation environment, the best construction method of internal rewards in MNAEK-PPO is determined, and a series of comparative experiments are carried out. The experimental results show that MNAEK-PPO greatly improves the exploration efficiency of decision space, and the convergence speed and game score are significantly improved, which provides a new solution for promoting the application and development of deep reinforcement learning in the generation of intelligent tactical strategies.
    Reference | Related Articles | Metrics
    Optimization of OpenMP Offload Shared Memory Access for Domestic Heterogeneous Platforms
    WANG Xin, LI Jianan, HAN Lin, ZHAO Rongcai, ZHOU Qiangwei
    Computer Engineering and Applications    2023, 59 (10): 75-85.   DOI: 10.3778/j.issn.1002-8331.2208-0017
    Abstract39)      PDF(pc) (1081KB)(34)       Save
    The local data share(LDS) on the heterogeneous processor DCU(deep computing unit) is an explicit addressable memory with low latency and high bandwidth. OpenMP of heterogeneous systems made in China does not provide the programming interface for LDS access, which leads to the ineffective use of LDS hardware to achieve efficient data access and storage. Aiming at this problem, the execution mode of OpenMP Offload for DCU platform, the allocation method of LDS and the instruction structure specific to LDS memory access are studied, and the manual support of LDS memory access is realized. In addition, aiming at the different execution modes of OpenMP Offload, the automation of LDS memory access is realized on the basis of this optimization method, and a set of efficient memory access strategies for domestic heterogeneous platforms is formed. The experiment is tested by using the standard test set of polybench. The average speedup of manual and automatic optimization methods is 2.60 in single-threaded mode, 1.38 in multi-threaded non-SPMD mode by manual optimization method and 1.11 in multi-threaded SPMD mode by automatic optimization method. The experimental results show that the automatic and manual support of LDS memory access is helpful to improve the running speed of OpenMP heterogeneous programs.
    Reference | Related Articles | Metrics
    Enhancing Molecular Topological Information with Multi-Task Graph Neural Networks
    JIANG Yelu, QUAN Lijun, WU Tingfang, LYU Qiang
    Computer Engineering and Applications    2023, 59 (10): 86-93.   DOI: 10.3778/j.issn.1002-8331.2208-0350
    Abstract48)      PDF(pc) (671KB)(33)       Save
    The predictions of molecular properties represented by molecular toxicities play an important role in the development of many fields mainly based on drug design, but it is always a challenge to quickly and accurately predict molecular toxicities by directly using the molecular structure information. At present, the emergence of deep learning methods such as convolutional networks and graph networks has made some progress in solving this problem. There are two key issues affecting the prediction performance of graph network-based deep learning methods in molecular toxicity prediction. Firstly, the data-driven nature makes the model still unreliable in the face of small data batches. Secondly, modeling the molecular structures only takes into account natural covalent bonds, which provides coarse-grained information. In order to solve the above problems, a novel way of molecular structure modeling, MT-ToxGNN, is presented. This method integrates the idea of multi-task into the graph neural network, which allows different tasks to learn the reliable distribution of different data from each other during training, thereby avoiding the problem of overfitting on small batch data. In addition, both intramolecular covalent bonds and non-covalent interactions are used to encode molecules into topological structures. That is, after constructing the edge sets of traditional graphs using molecular covalent bonds, the non-covalent interactions are used to construct the edge sets of novel graphs, thus compensating for the lack of molecular structure information represented by traditional graph. Then, the molecular covalent and non-covalent information is processed separately using specially designed graph networks to fully learn different molecular structures. In the performance comparison with a large number of state-of-the-art methods, MT-ToxGNN achieves the best Pearson coefficient metric on several molecular toxicity datasets.
    Reference | Related Articles | Metrics
    Adaptive Hybrid Strategy Sparrow Search Algorithm
    SU Yingying, WANG Shengxu
    Computer Engineering and Applications    2023, 59 (9): 75-85.   DOI: 10.3778/j.issn.1002-8331.2207-0081
    Abstract76)      PDF(pc) (966KB)(55)       Save
    A sparrow search algorithm based on adaptive hybrid strategy is proposed to solve the problems of low accuracy, insufficient stability and easy to fall into local optimization of sparrow search algorithm. The tent chaotic map is introduced to initialize the population, increase the population number, and merge the two populations. Then the elite population is obtained by using the elite strategy to improve the quality of the initial solution. The adaptive periodic convergence factor α is introduced to strengthen the search ability and convergence speed. The position update mode of followers and forerunners is adjusted to prevent the algorithm from falling into local optimization to a certain extent. Polynomial mutation disturbance is introduced to solve the problem of falling into local optimization of SSA. Using 12 test functions and the results show that AHSSSA has better optimization performance than SSA.
    Reference | Related Articles | Metrics
    Similarity-Based Training Set Recommendation for Software Defect Prediction
    WANG Chao, YU Qiao, HAN Hui
    Computer Engineering and Applications    2023, 59 (9): 86-94.   DOI: 10.3778/j.issn.1002-8331.2208-0297
    Abstract54)      PDF(pc) (823KB)(23)       Save
    In the process of software defect prediction, the quality of training set is the key factor affecting the prediction results. In recent years, training set selection has also become a research hotspot in cross-project defect prediction and cross-version defect prediction scenarios. However, most of the existing studies focus on a single prediction scenario, which may affect the quality of training set to some extent. Based on two scenarios of cross-project defect prediction and cross-version defect prediction, this paper proposes an approach of similarity-based training set recommendation(STSR) from the perspective of data distribution. Firstly, the candidate source projects and target project are divided into the same number of clusters by clustering, and the Euclidean distances between the cluster centers are calculated to measure the similarity of datasets. Then, the target project is sampled, and the defect rate differences between the candidate source projects and sampled target project are calculated. Moreover, the interference class ratio is calculated. Finally, the recommendation of training set is realized. The experiments are conducted on 40 versions of 11 projects from the PROMISE repository, and the performance of STSR is evaluated by F1 and AUC. The results indicate that the performance of STSR is better than that of cross-version defect prediction under F1, and similar under AUC. In terms of time cost, the maximum recommended time of STSR method is 5.09?s, which is acceptable.
    Reference | Related Articles | Metrics
    Robust Semi-Supervised Fuzzy C-Means Clustering for Time Series
    XU Jiucheng, HOU Qinchen, QU Kanglin, SUN Yuanhao, MENG Xiangru
    Computer Engineering and Applications    2023, 59 (8): 73-80.   DOI: 10.3778/j.issn.1002-8331.2207-0445
    Abstract87)      PDF(pc) (638KB)(48)       Save
    The fuzzy C-means clustering algorithm is sensitive to noisy data, and it fails to effectively utilize the supervised information contained in the small amount of labeled data in time series data. To address these problems, the paper proposes an improved robust semi-supervised fuzzy C-means clustering algorithm(SRFCM). Firstly, a sample uncertainty analysis method based on Mahalanobis distance is proposed, and add it to the semi-supervised fuzzy C-means clustering(SFCM) modeling to eliminate the influence of noise points. On this basis, by improving the partial supervision mechanism of SFCM, the supervision ability of labeled data is increased. And in the clustering process, the time warped edit distance(TWED), which can elastically measure the similarity of time series, is used instead of the traditional Euclidean distance. Through the experimental comparison of 7 groups of public time series datasets, the results show that the algorithm has excellent clustering effect.
    Reference | Related Articles | Metrics
    Design and Optimization of Query Operator on GPU
    LENG Fangling, LIU Jun, WU Yingying, BAO Yubin
    Computer Engineering and Applications    2023, 59 (8): 81-88.   DOI: 10.3778/j.issn.1002-8331.2208-0292
    Abstract81)      PDF(pc) (818KB)(21)       Save
    Selection, connection, projection and aggregation are the basic operations in traditional relational database. In order to realize the query optimization of relational database on GPU, the corresponding GPU algorithm must be used to realize the corresponding relational operator. Referring to the hierarchical design idea of divide and conquer of GDB, relational algebra is divided into operator layer and primitive layer. There are some difficult problems in the process of data query processing, such as data transmission delay, excessive use of shared memory, reduction of the number of active threads and communication delay caused by data communication between threads. To solve these problems, the query optimization algorithm is implemented based on the relatively new Pascal architecture. Based on the principle of the original connection, aggregation and condition selection algorithm, the corresponding algorithm is designed and optimized. The workload of each working thread is increased, the delay hiding between kernel computing and data transmission is realized, and the problem of data skew in connection operation is solved.
    Reference | Related Articles | Metrics
    Research onTwo-Stage Search Strategy for Constrained Many-Objective Optimization
    GENG Huantong, ZHOU Zhengli, SHEN Junye, SONG Feifei
    Computer Engineering and Applications    2023, 59 (7): 80-91.   DOI: 10.3778/j.issn.1002-8331.2207-0167
    Abstract65)      PDF(pc) (1069KB)(29)       Save
    In dealing with constrained many-objective optimization problems, a key issue of evolutionary algorithms is constraint handling and the tradeoffs between convergence anddiversity. However, the constraints in the search space hinder the population from finding the Pareto front, which tends to make the population fall into a local optimum, while the discrete feasible areas make the population less diverse. Therefore, a two-stage search strategywith the combined operator (TSCO) is proposed. TSCO deals with the constraints in two stages. Firstly, the algorithm only optimizes the objective function and the population is not constrained to approach the Pareto front direction rapidly. Secondly, the constraint violation degree is treated as a new objective function to solve the original constraint problem by objective transformation. A combined operator consisting of the simulated binary crossover operator and the DE/current-to-pbest/1 operator is used in the search process to generate individuals with excellent convergence and diversity. To verify the strategy effectiveness, AGE-MOEA combined with TSCO(TSCOEA) is compared with four state-of-the-art constrained many-objective evolutionary algorithms on the C_DTLZ, DC_DTLZ, and MW test suites. Experiments show that TSCOEA obtains better population convergence and diversity on most problems.
    Reference | Related Articles | Metrics
    Path Planning Method for Comprehensive Wild Terrain Based on Riemannian Manifold
    WANG Shuai, LIU Xiangyang
    Computer Engineering and Applications    2023, 59 (7): 92-101.   DOI: 10.3778/j.issn.1002-8331.2207-0323
    Abstract52)      PDF(pc) (890KB)(34)       Save
    It is of great practical significance to realize the path planning on the comprehensive wild terrain. It needs to consider various travel resistances such as elevation, land cover classification, vegetation density and wind direction. For the global path planning problem under such multi-resistance elements, the current solution cannot give a unified mathematical expression for this kind of problem due to the limitation of the Euclidean distance space. This makes the solution system chaotic, especially under the coexistence of isotropic and anisotropic resistances. Viewing the Earth’s surface from the perspective of a Riemannian manifold, the Riemannian metric that reflects local information and the generalized geodesic distance(comprehensive pathfinding cost) that helps form a distance spacecan be derived, thus giving a generalization for such problems. The unified mathematical expression reflects the actual situation and needs of field travelprecisely. Based on this, the improved heat method is used to solve the forest fire escape path planning problem as an application example of path planning on the comprehensive wild terrain.
    Reference | Related Articles | Metrics
    E-CenterNet Algorithm with Improved Multi-Scale Convolution Structure and Gaussian Kernel
    HU Songsong, WU Lianghong, ZHANG Hongqiang, CHEN Liang, ZHOU Bowen, ZHANG Lyu
    Computer Engineering and Applications    2023, 59 (6): 70-80.   DOI: 10.3778/j.issn.1002-8331.2205-0430
    Abstract94)      PDF(pc) (851KB)(66)       Save
    As the detection algorithm, the CenterNet based on ResNet and DLA(deep layer aggregation) as the backbone doesn’t have sufficient capabilities for feature extractions, neither its heatmap accords with the target real bounding box nor the keypoint loss function fully consider the impact of the predicted value on the training proportion of difficult and easy samples. In order to solve these deficiencies, E-CenterNet with more advanced multi-scale convolution structure and Gaussian kernel is proposed. Firstly, lightweight EfficientNetV2-S is introduced as the backbone, and the multi-scale convolution structure based on the pyramid split attention network is used to enhance the feature extraction ability. Secondly, by improving the Gaussian kernel, the heatmap generated by CenterNet is enhanced from a fixed circle to an ellipse varying with the width and height of the bounding box, which enhances the detection ability of algorithm for objects with large differences in the width and height of the bounding box. Finally, the Keypoint loss function based on the predicted value is proposed in order to augment the training ratio of the algorithm E-CenterNet for difficult samples. The experimental results on the Pascal VOC indicate that the mAP of the E-CenterNet accounts for 83.3%, which is 2.6 percentage points higher than that of the CenterNet, and the detection performance of E-CenterNet is better than that of the CenterNet.
    Reference | Related Articles | Metrics
    Chaotic Sparrow Search Algorithm Based on Multi-Directional Learning
    CHAI Yan, SUN Xiaoxiao, REN Sheng
    Computer Engineering and Applications    2023, 59 (6): 81-91.   DOI: 10.3778/j.issn.1002-8331.2205-0528
    Abstract97)      PDF(pc) (1268KB)(64)       Save
    Sparrow search algorithm in the iterative process has poor performance such as the low population diversity and the tendency of falling into local optima. To address the mentioned problems, a chaotic sparrow search algorithm based on multi-directional learning is proposed in this paper. The Hénon chaotic sequence is employed to initialize the population, the proposed initialization method increases the diversity of the sparrow population, and expands the search range of feasible solutions, which can lay the foundation for global optimization. The multi-directional learning strategy is adopted to increase the opportunity of sparrow followers to explore unknown areas, and to balance the local development performance and global search ability of the algorithm. Moreover, the mutation strategy of genetic algorithm is used to disturb and mutate the current optimal individual according to the dynamic mutation probability while the algorithm trapped in the local optimum. The MSSA algorithm is applied to the wireless sensor network node coverage optimization problem. The numerical experiment results and the Wilcoxon rank sum test results show that the MSSA algorithm has more obvious advantages in terms of convergence accuracy and convergence speed.
    Reference | Related Articles | Metrics
    Prediction of Disease Gene Based on Fusion Features and GA-SVM Algorithm
    TAN Zhuokun, LUO Longfei, WANG Shunfang
    Computer Engineering and Applications    2023, 59 (5): 70-77.   DOI: 10.3778/j.issn.1002-8331.2206-0472
    Abstract71)      PDF(pc) (607KB)(49)       Save
    The feature information provided by a single biological data network is limited. Aiming at this problem, a multi-network feature fusion method based on semi-supervised autoencoder is proposed to enrich feature information. In addition, in order to solve the problem of artificially setting the hyperparameters of the model, it is easy to cause problems such as low model performance and falling into local optimum, it is further proposed to use the genetic algorithm to optimize the support vector machine(GA-SVM algorithm) to improve the predictive performance of brain disease genes. First, the similarity data networks from different data sources are constructed, then the features are extracted from the four data networks by using the random walk with restart algorithm, and processed and fused by semi-supervised autoencoder, finally, under the strategy of 10-fold cross validation, GA-SVM algorithm model is used to predict disease genes, and compared with other algorithms. The experimental results show that the AUC and AUPR values on the PD dataset are 0.805 and 0.792, respectively, while the AUC and AUPR values on the MDD dataset are 0.825 and 0.823, respectively, which are superior to the existing models. It is proved that this method can effectively improve the prediction effect of brain disease genes.
    Reference | Related Articles | Metrics
    Multi-Strategy Particle Swarm Optimization Algorithm Based on Evolution Ability
    WANG Xiaoyan, CAO Dexin
    Computer Engineering and Applications    2023, 59 (5): 78-86.   DOI: 10.3778/j.issn.1002-8331.2111-0423
    Abstract87)      PDF(pc) (655KB)(62)       Save
    Aiming at the shortcomings of particle swarm optimization algorithm, such as easy premature convergence and low solution accuracy, a multi-strategy particle swarm optimization algorithm based on evolution ability is proposed. According to the change direction of fitness value, particles are divided into progressive particles and retrogressive particles. The progressive particles are updated according to the original evolution strategy, retaining the advantages of the original algorithm. For the retrogressive particles, it is further divided into temporarily retrogressive particles and long-term retrogressive particles according to the particle activity. For temporarily retrogressive particles, the dependence on individual historical speed is reduced or even learning in the opposite direction. For the long-term retrogressive particles, different evolution strategies are adopted according to the fitness value of the particles to improve the global optimization ability. At the same time, it designs a kind of inertia weight with random fluctuations, so that the particles still have the ability to jump out of the current area in the later stage of the algorithm, which is conducive to the global search. Comparing the optimization results with other algorithms in 10 test functions in different dimensions shows that this algorithm has advantages in both the convergence speed and accuracy of solving low-dimensional and high-dimensional problems. The EAMSPSO algorithm is applied to solve semi-infinite programming problems. Experimental results show that this algorithm is suitable for solving semi-infinite programming problems and has advantages.
    Reference | Related Articles | Metrics
    Ant Colony Algorithm Based on Dynamic Pheromone Update and Path Rewards and Punishments
    MA Shixuan, YOU Xiaoming, LIU Sheng
    Computer Engineering and Applications    2023, 59 (4): 64-76.   DOI: 10.3778/j.issn.1002-8331.2207-0340
    Abstract87)      PDF(pc) (818KB)(45)       Save
    Aiming at the fact that the ant colony algorithm is prone to fall into local optimum, the convergence speed is slow, and it is difficult to solve large-scale problems, this paper proposes an ant colony algorithm based on dynamic pheromone update strategy dependent on the information entropy and the number of stagnation  and the reward and punishment strategy based on the optimal path set. In the dynamic pheromone update strategy, the convergence coefficient is used to dynamically adjust the pheromone, so as to effectively balance the diversity and convergence of the algorithm. During the search process, the convergence speed is accelerated by continuously increasing the convergence coefficient. When the information entropy decreases or the number of stagnation reaches a certain value, the local optimum can be jumped out by reducing the convergence coefficient. At the same time, based on the optimal path set, the optimal path is rewarded and other paths are penalized, thereby reducing the number of cities that ants can choose at each step, and speeding up the convergence speed. And three local optimization methods are used to further improve the accuracy of the solution. After experimental tests, the algorithm is used to solve the traveling salesman problem(TSP), which has a high solution accuracy and can effectively balance the contradiction between the solution accuracy and the convergence speed.
    Reference | Related Articles | Metrics
    Novel Lichtenberg Algorithm Combining Partition-Oriented Search and Adaptive Diffusion
    LI Yongyu, MA Liang, LIU Yong
    Computer Engineering and Applications    2023, 59 (4): 77-88.   DOI: 10.3778/j.issn.1002-8331.2205-0566
    Abstract77)      PDF(pc) (834KB)(35)       Save
    Aiming at the problems of slow convergence and easy to fall into local optimization, a novel Lichtenberg algorithm(NLA) combining partition-oriented search and adaptive diffusion is proposed. According to the fitness value of the group particles, the search space is divided into the central area and the edge area, and the positions of the particles in the central area and the edge area are updated by using the dynamic tendency of the spiral coefficient and the randomness of the Levy variation, so as to increase the population diversity and strengthen the global search ability of the algorithm. An adaptive diffusion strategy is introduced, which makes full use of the position and fitness value information of each particle in the group to guide them to exchange information, avoid the algorithm falling into local extremums, and improve the local optimization ability of the algorithm. Using CEC2021 test function and 20 high-dimensional test functions with different characteristics for numerical experiments, and comparing the NLA algorithm with six different types of intelligent optimization algorithms, the experimental results show that the NLA algorithm has higher optimization accuracy and convergence speed. Finally, the effectiveness of the two improvement strategies on the NLA algorithm is verified.
    Reference | Related Articles | Metrics
    New Equilibrium Optimizer Algorithm Combining Concentration Equilibrium and Fick’s Law
    ZHANG Mengxi, MA Liang, LIU Yong
    Computer Engineering and Applications    2023, 59 (3): 66-76.   DOI: 10.3778/j.issn.1002-8331.2205-0007
    Abstract82)      PDF(pc) (722KB)(44)       Save
    New equilibrium optimizer algorithm(NEO), which combines concentration equilibrium and Fick’s law, is proposed to solve the problems of slow convergence speed, insufficient accuracy and information imbalance in development and search stage of equilibrium optimizer algorithm(EO). According to Brownian motion and diffusion phenomenon, different concentration balancing mechanism is adopted for particles in different concentration regions, and the algorithm balancing pool is improved to improve the information communication ability between populations. Then, two adaptive factors, power function and exponential function, are introduced into the algorithm parameters to further balance the global search and local development capabilities, so that the particle population can conduct extensive search and in-depth mining in the solution space. Finally, according to Fick’s law, a disturbance mechanism is introduced into the particle position updating formula to improve optimization accuracy and convergence speed of the algorithm. In addition, 24 benchmark test functions and Wilcoxon rank sum test are used to compare NEO algorithm with other intelligent optimization algorithms. The results show that NEO algorithm has good optimization performance.
    Reference | Related Articles | Metrics
    Research on Cascaded Labeling Framework for Relation Extraction with Self-Attention
    XIAO Lizhong, ZANG Zhongxing, SONG Saisai
    Computer Engineering and Applications    2023, 59 (3): 77-83.   DOI: 10.3778/j.issn.1002-8331.2205-0325
    Abstract92)      PDF(pc) (607KB)(39)       Save
    Named entity recognition and relation extraction are two essential subtasks in natural language processing and knowledge graph construction. In order to solve the problems of error transmission and entity sharing in the process of relation extraction, an entity relation extraction cascading labeling framework, Att-CasRel, which incorporates self-attention mechanism, is proposed. It can not only solve the cascading errors, but also solve the problem of multiple relational triples sharing the same entity in the same sentence. Based on Bert model, CB-Bert suitable for Chinese medical field is obtained by retraining text of CMeIE data set. Self-attention mechanism is incorporated in tail entity recognition stage to enhance feature expression of encoding vector of head entity and improve feature extraction capability of the model. Experimental results on CMeIE data set show that the proposed framework achieves better performance than independent extraction model extraction and other joint extraction models.
    Reference | Related Articles | Metrics
    Chaotic Tunicate Swarm Algorithm Based on Cosine Adaptive
    LI Xiangzhe, GU Lei, MA Li, WANG Mengjie
    Computer Engineering and Applications    2023, 59 (2): 65-75.   DOI: 10.3778/j.issn.1002-8331.2108-0237
    Abstract82)      PDF(pc) (5194KB)(58)       Save
    Aiming at the shortcoming of tunicate swarm algorithm, such as easy to fall into local optimum and slow convergence speed, a chaotic tunicate swarm algorithm based on cosine adaptive is proposed in this paper. In the simulation of the capsule jet propulsion behavior, the cosine adaptive curve is introduced to calculate the social forces between individuals, so that the improved algorithm is prone to prematurity. In addition, a chaotic behavior is added when the search individual moves to the optimal position, which makes the search individual avoid local optimum and has faster convergence speed. Finally, a variety of standard test functions are used to test, and the experimental results show that the new encapsulated swarm optimization algorithm proposed in this paper has better convergence speed, precision and global optimality while retaining the advantages of the original algorithm.
    Reference | Related Articles | Metrics
    Cross-Social Network User Matching Based on User Check-in
    DAI Jun, MA Qiang
    Computer Engineering and Applications    2023, 59 (2): 76-84.   DOI: 10.3778/j.issn.1002-8331.2203-0581
    Abstract61)      PDF(pc) (28513KB)(112)       Save
    Cross-social network user matching technology can integrate multi-platform user data to realize more diverse applications. Existing research on social network user matching based on check-in ignores the imbalance of multi-source social network check-in data, which leads to a decrease of matching accuracy under real datasets. Aiming at this problem, this paper proposes a cross-social network user matching method based on user check-in. Firstly, the user check-in data is coarse-grained and filtered through grid clustering algorithm, and the check-in data with strong potential correlation is selected; then the spatiotemporal features are extracted from the check-in data, and the similarity of different attributes is calculated; finally, by optimizing the multi-attribute weight distribution of similarity, comprehensive calculation of user matching score is conducted. Experimental results on multiple datasets demonstrate the effectiveness of the proposed method in the case of unbalanced check-in data.
    Reference | Related Articles | Metrics
    Adaptively Efficient Deep Cross-Modal Hash Retrieval Based on Incremental Learning
    ZHOU Kun, XU Liming, ZHENG Bochuan, XIE Yicai
    Computer Engineering and Applications    2023, 59 (2): 85-93.   DOI: 10.3778/j.issn.1002-8331.2108-0349
    Abstract57)      PDF(pc) (3556KB)(37)       Save
    To address the problems that current deep learning-based cross-modal hash retrieval algorithms cannot retrieve new category data and sub-optimal solution caused by relaxing discretization constraint of hash codes, an adaptive deep incremental hashing(ADIH) retrieval algorithm is proposed to directly learn the hash codes of newly coming data meanwhile keeping the old trained data unchanged. In order to preserve the similarity and dissimilarity among multi-modal data, hash codes will be projected into latent semantic space where binary constrained discrete cross-modal hash algorithm is introduced to optimize hash code without using any relaxation. Besides, considering that there is currently no effective method which can be used to evaluate complexity of deep hashing methods, a novel method based on neuron updating operation is proposed to analyze the complexity. The experimental results on the public datasets show that the training time of the proposed algorithm is much lower than that of the comparison algorithms, and the retrieval accuracy is higher than that of the comparisons.
    Reference | Related Articles | Metrics
    Evolutionary Algorithm Based on Niche for Multi-Objective Optimization
    GU Qinghua, LUO Jiale, LI Xuexian
    Computer Engineering and Applications    2023, 59 (1): 126-139.   DOI: 10.3778/j.issn.1002-8331.2207-0006
    Abstract81)      PDF(pc) (1362KB)(45)       Save
    The main challenges for evolutionary algorithms to balance convergence and diversity in solving many-objective optimization problems are in two aspects:strengthening selection pressure towards the Pareto optimal front and achieving a good diversity of the obtained solution set. However, the Pareto dominance-based selection criteria cannot effectively solve the above problems with the increase of the objective number. Therefore, an evolutionary algorithm based on niche is designed. Firstly, based on the niche, a new dominance relation is proposed, where an aggregation function and an objective vector angle-based density estimation method are used to measure the convergence and distribution of candidate solutions respectively. Then, in the same niche, only the solution with the best convergence degree is identified as the non-dominated solution to ensure the convergence of a solution set. At the same time, to improve the diversity of a solution set, in any two different niches, the solution with good convergence and distribution in one niche will dominate the solution with poor convergence and distribution in the other niche. Finally, the proposed dominance relation is embedded into VaEA instead of Pareto dominance relation to design an improved many-objective evolutionary algorithm(VaEA-SDN). VaEA-SDN and six state-of-the-art algorithms NSGA-III, VaEA, MSEA, NSGAII-CSDR, RPS-NSGAII and CDR-MOEA are conducted in simulation experiments on the DTLZ(Deb-Thiele-Laumanns-Zitzler) and MaF(many-objective function) benchmark suites. The results show that the ability of VaEA-SDN in keeping a balance between the convergence and diversity has an average improvement of 37.7%, 32.9%, 31.8%, 22.2%, 43.5%, 30.2% over the compared six algorithms in terms of the quality of obtained solutions.
    Reference | Related Articles | Metrics
    Improved Robust Fuzzy Twin Support Vector Machine Algorithm
    ZHOU Yuqun, ZHANG Desheng, ZHANG Xiao
    Computer Engineering and Applications    2023, 59 (1): 140-148.   DOI: 10.3778/j.issn.1002-8331.2204-0496
    Abstract62)      PDF(pc) (1923KB)(44)       Save
    Since the fuzzy twin support vector machine(FTSVM) algorithm is still sensitive to noise and prone to over fitting, as well as cannot effectively distinguish support vectors from outliers. This paper proposes an improved robust fuzzy twin support vector machine(IRFTSVM). Firstly, a new kind of mixed membership function is constructed by combining the intra-class hyperplanemembership function and the improved [k]-nearest neighbor membership function. Secondly, a regularization term and the additional constraint are brought into the objective function to minimize the structural risk, avoid the computationof inverse matrix, and nonlinear problems can be directed from linear case as the classical SVM algorithm. Finally, the hinge loss function is replaced by the pinball loss function to reduce the noise sensitivity. In addition, the proposed algorithm is assessed and compared with SVM, TWSVM, FTSVM, PTSVM and TBSVM on some UCI datasets and an artificial dataset. The experimental results show that the proposed algorithm is satisfactory.
    Reference | Related Articles | Metrics
    Arithmetic Optimization Algorithm Assisted by Support Vector Machine and Its Application
    TIAN Lu, LIU Sheng
    Computer Engineering and Applications    2022, 58 (24): 73-82.   DOI: 10.3778/j.issn.1002-8331.2205-0331
    Abstract64)      PDF(pc) (1855KB)(63)       Save
    Aiming at the shortcomings of arithmetic optimization algorithm, such as poor population diversity and easily into the local optimal solution, an improved arithmetic optimization algorithm assisted by support vector machine is proposed. First of all, the concept of balance pool in the balance optimizer algorithm is proposed. The balanced pool brings together descendant and average candidate solutions generated based on four mutational strategies in success-history based adaptive DE algorithm. The strategy is used to improve the diversity of population. Secondly, the support vector machine algorithm is introduced to calculate the individual retention rate by integrating individual fitness value and distance between individuals. SVM is used to classify the candidate solutions in the balance pool, and only the dominant candidate solutions are reserved. Then, the dominant candidate solutions are sorted according to the retention rate, and the first [N] individuals are reserved to the next generation to build a new balance pool. Finally, the simulation results of SVMAOA and other optimization algorithms on the benchmark function show that the improved algorithm has higher searching accuracy and faster convergence speed. The feature selection method based on SVMAOA is tested in seven UCI data sets. By evaluating the average classification accuracy and the number of selected features, it is concluded that the algorithm can effectively reduce the feature dimension and achieve data classification, which has certain engineering application value.
    Reference | Related Articles | Metrics
    Clustering of Single-Cell RNA-Seq Data Based on Heterogeneous Parallel Computing
    XIE Linjuan, LI Lixuan, ZHANG Shaoqiang
    Computer Engineering and Applications    2022, 58 (24): 83-89.   DOI: 10.3778/j.issn.1002-8331.2204-0340
    Abstract62)      PDF(pc) (2701KB)(60)       Save
    With the development of single-cell RNA sequencing(scRNA-seq) technology, the mainstream scRNA-seq throughput has grown from thousands of cells to tens of thousands of cells. Cell typing based on scRNA-seq data is one of the important problems in cell research, which mainly uses unsupervised clustering methods. The existing clustering methods for large-scale single-cell sequencing data reduce the time complexity by simplifying the single-cell network, which leads to the accuracy decline of cell typing. However, the common cell typing methods with high accuracy cannot handle large-scale data. For this reason, this study adopts the combination of [k]-nearest neighbors(KNN) and cell-cell similarity threshold to construct a new single-cell network, uses CPU+GPU heterogeneous parallel computing to improve the computing speed, and finally performs cell clustering by an improved Markov clustering algorithm. Through experiments on seven large-scale single-cell datasets, it is found that the algorithm has better clustering accuracy than the main algorithms, and thus is suitable for cell typing of scRNA-seq data produced by mainstream technologies.
    Reference | Related Articles | Metrics
    Cross-Domain Recommendation Algorithm Based on Deep Fusion of Side Information
    LU Yongqian, SHENG Jiagen
    Computer Engineering and Applications    2022, 58 (24): 90-96.   DOI: 10.3778/j.issn.1002-8331.2107-0493
    Abstract37)      PDF(pc) (1026KB)(30)       Save
    Collaborative filtering has been successfully used to provide users with personalized products and services. However, it faces the data sparsity problem as well as the cold start problem. One solution is to incorporate side information and the other is to learn knowledge from related fields. In this paper, a cross-domain recommendation algorithm based on deep fusion of side information CICDR is proposed by considering both, which integrates collective matrix factorization and deep transfer learning. The algorithm uses Semi-SDAE and matrix factorization(MF) to model in both source and target domain to learn the effective feature vectors fromrating information and side information, and makes more accurate recommendation by using the user’s implicit feedback information. In this way, the user and project potential factors learned in the two fields retain more semantic information for recommendation. Then, the incomplete orthogonal nonnegative matrix tri-factorization(IONMTF) is used to generate a common potential factor bridging the two related fields to alleviate the cold start problem and data sparsity problem in the target domain. The comparison with four classic algorithms on three real data sets verifies the effectiveness of the proposed algorithm and further improves the recommendation accuracy and user satisfaction.
    Reference | Related Articles | Metrics
    Collaborative Filtering Deep Recommendation Algorithm Based on Time Feature
    WEI Ziyu, ZHU Xiaodong, XU Yi
    Computer Engineering and Applications    2022, 58 (23): 67-73.   DOI: 10.3778/j.issn.1002-8331.2106-0434
    Abstract76)      PDF(pc) (965KB)(67)       Save
    Aiming at the problem of data sparsity and cold start in recommendation algorithm, this paper proposes a convolutional neural network based collaborative filtering depth recommendation algorithm(CNN deep recommendation algorithm with time, C-DRAWT) and a multi-layer perceptron based collaborative filtering depth recommendation algorithm(MLP deep recommendation algorithm with time, M-DRAWT). The algorithm first performs data preprocessing, and uses binary to encode user and project information, which alleviates the sparseness of books in one-hot encoding. Subsequently, the hidden features of the user and the item are extracted, and the features of the user and the item are combined with the timestamp feature, and then input into the optimized convolutional neural network and the multi-layer perceptron respectively, and finally the recommended item at the latest time is obtained. The two algorithms are verified by comparative experiments based on MovieLens-1M data set. The F1 score value and RMSE value are increased by 0.78% and 2.7% respectively. The results show that this method can alleviate the problems of data sparsity and cold start, and has better recommendation effect than the previous model.
    Reference | Related Articles | Metrics
    Improved Grey Wolf Optimization Algorithm Based on Levy Flight and Dynamic Weight Strategy
    DING Ruicheng, ZHOU Yucheng
    Computer Engineering and Applications    2022, 58 (23): 74-82.   DOI: 10.3778/j.issn.1002-8331.2205-0577
    Abstract89)      PDF(pc) (1026KB)(58)       Save
    Aiming at the disadvantages of grey wolf optimization(GWO), such as insufficient global search ability and easy to fall into local optimization in complex optimization problems, an improved grey wolf optimization(LGWO) with Levy flight and dynamic weight strategy is proposed. Firstly, Singer mapping is used to initialization to increase the diversity of the population. Secondly, a new nonlinear convergence factor updating strategy is adopted to balance the global and local search abilities. Finally, Levy flight and dynamic weight strategy are introduced into the position update formula to reduce the risk of falling into local optimum and improve the optimization accuracy. The performance of algorithm is evaluated by comparing experiments on 8 benchmark functions with other optimization algorithms and improved algorithms. The experimental results indicate that the LGWO algorithm is superior to other algorithms in convergence speed and prediction accuracy and the validity of the LGWO algorithm in high dimensional problems is verified.
    Reference | Related Articles | Metrics
    Research on Test Data Generation for Improved Chimpanzee Optimization Algorithm
    GAO Dahuan, LIANG Hongtao, DU Junwei, YU Xu, HU Qiang
    Computer Engineering and Applications    2022, 58 (23): 83-93.   DOI: 10.3778/j.issn.1002-8331.2204-0228
    Abstract58)      PDF(pc) (871KB)(42)       Save
    The key of automatically generating test data is whether it can generate data with high coverage and strong error-correcting ability. In order to solve the problems of low test data generation efficiency and low convergence precision of chimp optimization algorithm, a chimp optimization algorithm based on sine-cosine perturbation strategy(SC-ChOA) is proposed. Firstly, the Latin hypercube strategy is used to initialize the population and enhance the diversity of the population. Secondly, the nonlinear decay convergence factor is introduced to balance the global and local exploration ability of the algorithm, to avoid the algorithm stagnation caused by the local range search of the group. In addition, the test function is compared with standard chimp optimization algorithm and common genetic algorithm to verify the effectiveness of the algorithm. Finally, the improved algorithm is applied to test data generation, in order to optimize the test data, a branch function is inserted into the pile to establish the fitness function. In order to verify the effectiveness of the improved algorithm in test data generation, several benchmark programs are used to compare the algorithms, the results show that SC-ChOA has obvious advantages in the coverage, average iteration times and running time of test data generation.
    Reference | Related Articles | Metrics
    Multi-Feature Relationship Classification Algorithm Fused with Fine-Grained Entity Types
    ZUO Yayao, YI Biao, LI Wenjie
    Computer Engineering and Applications    2022, 58 (22): 65-71.   DOI: 10.3778/j.issn.1002-8331.2106-0278
    Abstract81)      PDF(pc) (756KB)(73)       Save
    In recent years, relationship classification based on deep learning has mostly focused on the improvement of attention mechanism or the optimization of semantic information, but the features extracted by such methods are often relatively single, ignoring the impact of entity types on relationship classification, and there is semantic information problem such as incomplete learning. This paper proposes a new relationship classification method Type-SBNE, which introduces fine-grained entity type information for entity type learning tasks, and generates entity class vectors by averaging all entity vectors in each entity type, and then the corresponding feature vectors are obtained through entity information learning and sentence meaning information learning, and composite semantic features are obtained through splicing and fusion. Finally, the fully connected layer and Softmax function are used to predict the relationship between entity pairs. Based on the fine-grained entity type information, the characteristics of the entity are enriched, and the expression of each entity in the context is effectively strengthened. Experiments show that the Type-SBNE model can better complete the relationship classification task, and the effect is improved compared with other algorithms.
    Reference | Related Articles | Metrics
    Deep Collaborative Filtering Algorithm Based on Multi-Dimensional Feature Crossover
    LU Yuecong, WANG Ruiqin, JIN Nan
    Computer Engineering and Applications    2022, 58 (22): 72-78.   DOI: 10.3778/j.issn.1002-8331.2105-0306
    Abstract78)      PDF(pc) (886KB)(70)       Save
    The recommendation algorithm based on deep learning initially takes the ID information of user and item as input. However, the ID cannot well show the characteristics of user and item. In the original data, the user’s rating data on the item can show the characteristics of user and item to a certain extent. In this paper, implicit feedback and ID information are used as features of users and items in the rating task, which can eliminate the noise caused by users’ subjectivity and alleviate the problem of cold start to a certain extent. Single layer neural network is used to reduce the dimension of original high-dimensional sparse features, and feature crossover is used to get the low-order interaction between users and items, and then neural network is used to capture the high-order interaction between users and items, The high-order and low-order interactions between features are effectively extracted. Experiments on four public data sets show that the algorithm proposed in this paper can effectively improve the recommendation accuracy.
    Reference | Related Articles | Metrics
    Oversampling Algorithm Based on Density Peak Clustering and Radial Basis Function
    LU Miaofang, YANG Youlong
    Computer Engineering and Applications    2022, 58 (21): 67-74.   DOI: 10.3778/j.issn.1002-8331.2103-0564
    Abstract63)      PDF(pc) (974KB)(35)       Save
    Most of the existing oversampling algorithms only consider the distribution of the minority instances but ignore the distribution of the majority instances in the sampling process. In addition to the problem of imbalance between classes, the data set also has the problem of imbalance within classes. To solve these problems, this paper proposes a new oversampling method based on density peak clustering and radial basis function. Firstly, the minority instances are adaptively clustered by the improved density peak clustering algorithm, and a number of minority sub-clusters are obtained. Secondly, the local density calculated by the clustering process is used to assign weights to each sub-cluster, which are used to determine the required number of each sub-cluster. Finally, the radial basis function is used to calculate the mutual minority class potential of each minority instances, and the minority class is oversampled based on the mutual minority class potential. The proposed algorithm is combined with different classifiers to conduct experiments, and different indicators are used to evaluate the performance. The experiment shows that the performance of the proposed algorithm is better.
    Reference | Related Articles | Metrics
    Density Peak Clustering Algorithm Combining Density-Ratio and System Evolution
    CAO Junrong, ZHANG Desheng, XIAO Yanting
    Computer Engineering and Applications    2022, 58 (21): 75-82.   DOI: 10.3778/j.issn.1002-8331.2105-0444
    Abstract64)      PDF(pc) (789KB)(34)       Save
    The density peak clustering(DPC) algorithm can effectively cluster non-spherical data. However, the algorithm needs to input the cutoff distance and manually intercept the clustering center, which causes the poor clustering effect of the DPC algorithm sometimes. To solve these problems, this paper proposes a density peak clustering algorithm combining density-ratio and system evolution(DS-DPC). The natural nearest neighbor search is used to obtain the number of neighbors of each sample point, and the density calculation formula is improved according to the idea of density-ratio, so that it can reflect the distribution of surrounding samples. According to the ranking value, the product of the local density and the relative distance is sorted in descending order, the cluster centers, and clusters the remaining samples are selected, according to the allocation strategy of the DPC algorithm, avoiding the subjectivity of manually selecting the cluster centers. The system evolution method is used to determine whether the clustering results need to be merged or separated. Through experiments on multiple data sets and comparison with other clustering algorithms, the experimental results show that the proposed algorithm has a better clustering effect.
    Reference | Related Articles | Metrics
    Coefficient Enhanced Least Square Regression Subspace Clustering Method
    JIAN Cairen, WENG Qian, XIA Jingbo
    Computer Engineering and Applications    2022, 58 (20): 73-78.   DOI: 10.3778/j.issn.1002-8331.2103-0480
    Abstract78)      PDF(pc) (817KB)(43)       Save
    In view of the fact that the least square regression subspace clustering method ignores the similarity between samples when solving the representation coefficients, an improved method is proposed. Based on the representation coefficient matrix of the sample mutual reconstruction and the similarity matrix of the sample having a great correlation, it defines the coefficient enhancement term to solve the representation coefficient matrix that can preserve sample similarity. Experiments on 8 standard data sets show that the proposed method can improve the performance of least square regression subspace clustering method.
    Reference | Related Articles | Metrics
    Improved Particle Swarm Optimization Algorithm Based on Attraction-Repulsion and Bidirectional Learning Strategies
    WANG Yawen, QIAN Qian, FENG Yong, FU Yunfa
    Computer Engineering and Applications    2022, 58 (20): 79-86.   DOI: 10.3778/j.issn.1002-8331.2105-0253
    Abstract64)      PDF(pc) (681KB)(39)       Save
    An improved particle swarm optimization(PSO) algorithm which combines attraction-repulsion strategy and bidirectional learning strategy is proposed to overcome the shortcomings of traditional PSO, such as slow convergence speed and easy to fall into local extreme. The bidirectional learning strategy overcomes the limitations of the traditional one-way learning method by expanding the searching range of the particles, so as to enrich the diversity of the population. In the attraction-repulsion strategy, the particle can be guided by the global best and the global worst particles to evolve towards the better direction, which improves the local optimization performance and convergence ability of the algorithm. Furthermore, in order to overcome the problem that the single learning factor and inertial weight cannot adjust the optimization process well when optimizing complex functions, a dual adaptive strategy is proposed to better balance the search behavior of the particles in the group. Finally, the proposed algorithm is simulated and verified by using the standard test function, and compared with the other two improved algorithms. The experimental results show that the improved algorithm has advantages in the optimization effect and convergence speed under the same experimental conditions.
    Reference | Related Articles | Metrics
    Automatic Repair of C Program Memory Error Guided by Tracking Mechanism
    DONG Yukun, WEI Xinxin, SUN Yuxue, TANG Daolong
    Computer Engineering and Applications    2022, 58 (19): 76-87.   DOI: 10.3778/j.issn.1002-8331.2202-0254
    Abstract62)      PDF(pc) (861KB)(34)       Save
    C is a highly efficient and widely used language, but its widespread use has been accompanied by growing security issues. The memory error is a common error in C programs, which will cause the system to collapse in severe cases. Manually fixing memory error requires considerable efforts, and potentially introduces new errors in the repair process. To address this problem, an automatic program repair method based on tracking mechanism is proposed. Firstly, it constructs the scope-tree containing the distribution of variables in the program file. Then, a global pointer-based tracking mechanism is proposed to track the state of allocated memory with errors in the program. Finally, a patch is automatically generated based on a global pointer, and a scope-tree is used to locate where the defect is fixed so that the memory error can be safely repaired. Based on the above process, the prototype tool DTSFix has implemented, in addition evaluated it with open-source programs. The experimental results show that DTSFix can effectively detect and repair the real defects in open-source programs without side effects.
    Reference | Related Articles | Metrics
    Multi-Attribute Group Decision Making Algorithm with Probabilistic Dual Hesitant Fuzzy Sets and PROMETHEE Method
    ZHU Yuting, ZHANG Wenyu, HOU Junjie, ZHANG Renjie
    Computer Engineering and Applications    2022, 58 (19): 88-97.   DOI: 10.3778/j.issn.1002-8331.2203-0483
    Abstract51)      PDF(pc) (675KB)(32)       Save
    A multi-attribute group decision-making method based on probabilistic dual hesitant fuzzy sets and the PROMETHEE method is proposed to solve the multi-attribute group decision-making problem in the field of hesitant fuzzy information. Firstly, each decision expert’s probabilistic dual hesitant fuzzy information matrix is constructed. Secondly, the maximum dispersion method and the entropy value method are used to determine the objective weight of each decision-making expert and each index attribute, then the score function and the deviation function are combined to obtain the comprehensive decision-making evaluation information matrix of the decision-making experts. Then, the final decision results are obtained based on the method of combining probabilistic dual hesitant fuzzy sets with PROMETHEE. Finally, the method is applied to the analysis of the case analysis of the emergency response scheme for evaluating the aviation disaster accident, and compared with the calculation results of the TOPSIS, VIKOR, and PDHFS decision-making methods, the validity and reliability of the method have been verified.
    Reference | Related Articles | Metrics
    Link Prediction Algorithm Fusing Node Label and Strength Relationship
    WANG Shuyan, GONG Jingyi
    Computer Engineering and Applications    2022, 58 (18): 71-77.   DOI: 10.3778/j.issn.1002-8331.2204-0004
    Abstract97)      PDF(pc) (718KB)(43)       Save
    It is a difficult problem how to make link prediction through known paths combined with relevant attribute information and different relationship strengths. To solve this problem, a link prediction algorithm that fuses node labels and strong and weak relationships is proposed. Two central nodes are selected, and all network node labels centered on them are calculated by the algorithm based on the double radius node label. The [h]-depth local sub-graph with the node label of the central node is generated. the local sub-graph is extracted and used as the target network. The feature matrix is obtained, and then the feature matrix is matrix decomposing into the node attribute information and the strong and weak relationship, and dynamic weights are assigned to construct the similarity matrix. The experimental results show that compared with the common link prediction algorithms based on common neighbor algorithm and network embedding, the accuracy of this algorithm is improved by up to 1.83%, and the accuracy and efficiency of its prediction results are significantly improved. At the same time, it can be effectively and accurately mined the internal correlation of each node.
    Reference | Related Articles | Metrics
    Robust Least Squares Twin Support Vector Machine and Its Sparse Algorithm
    JIN Qifan, CHEN Li, XU Mingliang, JIANG Xiaoheng
    Computer Engineering and Applications    2022, 58 (18): 78-89.   DOI: 10.3778/j.issn.1002-8331.2203-0037
    Abstract70)      PDF(pc) (810KB)(46)       Save
    Least squares twin support vector machine(LSTSVM) solves two linear programming problems instead of solving complex quadratic programming problems. LSTSVM has the advantages of simple calculation and fast training speed. However, the hyperplanes obtained by LSTSVM are easily affected by outliers and the solution to LSTSVM lacks sparse. To solve this problem, a robust least squares twin support vector machine(R-LSTSVM) based on truncated least squares loss is proposed, and it verifies that the new model is robust to outliers in theory. Furthermore, in order to handle large-scale datasets, a sparse solution to the R-LSTSVM is obtained based on the representation theorem and incomplete Cholesky decomposition, and it proposes a sparse R-LSTSVM algorithm which is suitable for dealing with large-scale datasets with outliers. Numerical experiments show that compared with the existing algorithm, the classification accuracy, sparsity and convergence speed of the new algorithm are improved by 1.97%-37.7%, 26-199 times and 6.6-2 027.4 times, respectively.
    Reference | Related Articles | Metrics
    Success Fail History Based Hybrid RUNgeKutta Optimizer and Slime Mould Algorithm
    LIU Yusong, LIU Sheng
    Computer Engineering and Applications    2022, 58 (17): 61-71.   DOI: 10.3778/j.issn.1002-8331.2111-0546
    Abstract97)      PDF(pc) (912KB)(38)       Save
    A success fail history based hybrid RUNgeKutta optimizer and slime mould algorithm is proposed to solve the shortcoming of low convergence speed and low precision of the basic slime mould algorithm. Firstly, an improved success fail history archive mechanism for storing position information is proposed, and the change in individual fitness value is used as the probability of each archive record participating in calculations, and this mechanism is added to the original slime mould algorithm. Secondly, the RUNgeKutta optimizer and the improved slime mould algorithm are merged through parallel computing and information exchange to guide the slime mould algorithm to jump out of the local optimum, and to improve the slime mould algorithm in iterative when the accuracy of the solution in the narrow space. Thirdly, a communication strategy combining long and short time intervals is proposed to determine the timing of the communication between the two populations. Finally, a series of population information communication mechanisms based on space movement are proposed, without affecting two algorithms’ characteristics and advantages, the limitations of the two algorithms are overcome at the same time. For the experiment part, it uses the CEC2017 benchmark test functions, with traditional statistical indicators, MAE ranking, and Wilcoxon rank-sum test to verify the effectiveness of the new algorithm. At the same time, it explores high-dimensional functions and compares with the state-of-the-art swarm intelligence algorithms and improved algorithms in recent years. Also, the incomplete algorithms of this algorithm are compared and tested. The experimental results show that the improved strategy in this paper is effective and has portability, this algorithm’s accuracy and robustness are more competitive.
    Reference | Related Articles | Metrics