Computer Science

Contents

Computer Science. 2022, 49 (6): 0-0.

Abstract

PDF(5385KB) ( 695 )

RelatedCitation | Metrics

PPO Based Task Offloading Scheme in Aerial Reconfigurable Intelligent Surface-assisted Edge Computing

XIE Wan-cheng, LI Bin, DAI Yue-yue

Computer Science. 2022, 49 (6): 3-11. doi:10.11896/jsjkx.220100249

Abstract

PDF(2653KB) ( 1055 )

References | Related Articles | Metrics

In order to compensate the performance loss caused by obstacle blocking in mobile edge computing (MEC) system in 6G-enabled “intelligent Internet of Things”,this paper proposes a partial task offloading scheme supported by aerial reconfigurable intelligent surface (RIS).Firstly,we investigate the joint design of the RIS phase shift vector,the proportion of offloading task,time slot allocation,the transmit power of users and the position of UAV,formulating a non-convex problem for minimization of the total energy consumption of users.Then,the original non-convex problem is decomposed into four subproblems,and the proximal policy optimization (PPO) method in deep reinforcement learning (DRL) is utilized to provide time slot allocation.The alternative optimization (AO) is leveraged to decouple the original problem into four subproblems,including the RIS phase shift design,the convex optimization of transmit power and offloading task amount,and the UAV altitude optimization.Simulation results show that the proposed PPO model can be trained quickly,the total energy consumption of users can be reduced by about 23% and 5.3%,compared with the fully-offload strategy and fixed-UAV-height strategy,respectively.

Multi-Task and Multi-Step Computation Offloading in Ultra-dense IoT Networks

ZHOU Tian-qing, YUE Ya-li

Computer Science. 2022, 49 (6): 12-18. doi:10.11896/jsjkx.211200147

Abstract

PDF(2370KB) ( 875 )

References | Related Articles | Metrics

With the rapid development of Internet of Things(IoT),various IoT mobile devices(IMDs) need to process more and more computing-intensive and delay-sensitive tasks,which puts forward new challenges for the mobile edge networks.To address these challenges,the MEC-equipped ultra-dense IoT has emerged.In such networks,IMDs can save their computation resources and reduce their energy consumption by offloading computing-intensive tasks to edge computing servers for processing.However,it will result in additional transmission time and higher delay.In view of this,an optimization problem is formulated for finding the trade-off between energy consumption and delay,which jointly considers the user(IMD) association,computation offloading and resource allocation for ultra-dense MEC-enabled IoT.To further balance the network load and fully utilize the computation resources,the optimization problem is finally modeled as multi-step computation offloading one.At last,an intelligent algorithm,adaptive particle swarm optimization(PSO),is utilized to solve the proposed problem.Compared with traditional PSO,the total cost of adaptive PSO reduces by 20%~65%.

Performance Analysis on Reconfigurable Intelligent Surface Aided Two-way Internet of Things Communication System

DONG Dan-dan, SONG Kang

Computer Science. 2022, 49 (6): 19-24. doi:10.11896/jsjkx.220100064

Abstract

PDF(2608KB) ( 886 )

References | Related Articles | Metrics

Reconfigurable intelligent surface(RIS) can intelligently change the wireless propagation environment to significantly improve the performance of wireless communication systems.Compared with traditional relay systems,it has the characteristics of low cost,low power consumption and easy deployment.It is regarded as one of the potential key technologies of 6G.Since RIS can dynamically change the phase characteristics of radio waves,the scalability of the network can be achievedby adjusting the phase shift reasonablely,and massive IoT nodes in the network can be flexibly served.In order to further improve the performance of the RIS-assisted IoT transmission system,a two-way RIS-assisted transmission system is proposed.By introducing full-duplex and self-interference cancellation technology,the system capacity and transmission efficiency are effectively improved.The analy-tical expressions for the outage probability,average bit error rate and average channel capacity of the proposed system are derived,and the relationship between system performance and system parameters such as the number of RIS reflecting elements in the system is obtained.The accuracy of the derivation and the performance advantages of the proposed scheme have been verified by Monte Carlo simulation.

Study on Task Offloading Algorithm for Internet of Vehicles on Highway Based on 5G MillimeterWave Communication

QIU Xu, BIAN Hao-bu, WU Ming-xiao, ZHU Xiao-rong

Computer Science. 2022, 49 (6): 25-31. doi:10.11896/jsjkx.211100198

Abstract

PDF(2417KB) ( 629 )

References | Related Articles | Metrics

With the rapid development of the Internet of vehicles,the emerging new types of in-vehicle tasks put forward higher requirements for communication and computing capabilities.The development of satellite communication technology and the large-scale deployment of 5G millimeter-wave base stations provide safer and more reliable services for highway vehicle users.At the same time,mobile edge computing technology deploys mobile edge computing(MEC) servers with computing and storage capabi-lities around user terminals to provide computing services for on-board tasks while reducing transmission delays.Aiming at the problem of offloading decision-making and communication resource allocation of vehicle tasks in highway scenarios,the joint optimization problem of computing and communication resources is modeled as a 0-1 mixed integer linear programming problem.Firstly,the original optimization problem is decoupled into the resource block allocation sub-problem and the offloading decision sub-problem.Secondly,the sub-problems are solved by using the water injection algorithm and the particle swarm algorithm.Finally,the sub-problems are iteratively solved based on the heuristic algorithm to obtain the optimal resource block allocation scheme and offload decision vector.Simulation results show that the algorithm minimizes the average system delay while meeting the requirements of all on-board missions.

Blockchain Sharding and Incentive Mechanism for 6G Dependable Intelligence

WANG Si-ming, TAN Bei-hai, YU Rong

Computer Science. 2022, 49 (6): 32-38. doi:10.11896/jsjkx.220400004

Abstract

PDF(2887KB) ( 596 )

References | Related Articles | Metrics

The sixth generation(6G) wireless communication network will become the base of endogenous intelligence,ubiquitous connectivity,and full-scene interconnection.It is an important basis to realize dependable intelligence in the future.Blockchain is considered as the key decentralized-enabled technology to improve the performance of 6G networks.In the future,the consensus nodes of the blockchain will be composed of massive edge devices and connected through wireless networks.However,motivating self-interest edge devices to participate in the consensus process still faces the challenges of information asymmetry,resource constraints and heterogeneous wireless communication environment.To solve these challenges,a blockchain sharding framework and an incentive mechanism for trusted and dependable intelligence in 6G are proposed.Firstly,an incentive mechanism is presented based on contract theory,which aims to maximize the benefits and reliability of the blockchain sharding.By analyzing the practical byzantine fault tolerance (PBFT) based intrashard consensus mechanism,this paper design energy consumption model for auditing and transmitting the blocks in wireless networks.Secondly,in order to improve the system reliability,it proposes a reputation mechanism based on subjective logic.Finally,a set of optimal contracts under complete information and asymmetric information scnearios are abtained,which could optimize the block revenue for blockchain service requester,while ensuring some desired economic properties,i.e.,budget feasibility,individual rationality and incentive compatibility.Simulation results show that the proposed contract-based incentive mechanism can motivate edge devices to participate in the blockchain consensus process and maintain the operation of blockchain from the perspective of economics more efficiently.

Wireless Resource Allocation Algorithm with High Reliability and Low Delay for Railway Container

XU Hao, CAO Gui-jun, YAN Lu, LI Ke, WANG Zhen-hong

Computer Science. 2022, 49 (6): 39-43. doi:10.11896/jsjkx.211200143

Abstract

PDF(1905KB) ( 570 )

References | Related Articles | Metrics

The intelligent container system improves the container operation efficiency through the real-time collection and transmission of information.In order to ensure the ultra reliable and low delay communication of container terminals,this paper pro-poses to transmit the information in the form of the short packets,and studies the sum rate maximization of the uplink transmission system for the railway container terminals in the multi cells.This paper proposes a spectrum resource allocation problem.Multiple cells share spectrum resources,and the terminals in the cells obtain the spectrum through competition.The game theory model is used to construct this competition relationship,and Nash equilibrium solution is proved.The best Nash equilibrium solution is the global optimal solution for system and rate maximization.Then,a distributed iterative algorithm is designed,which only needs local information interaction.It is proved theoretically that when the smoothing coefficient is small enough,the algorithm can converge to the best Nash equilibrium point with any high probability.Finally,the proposed algorithm is verified by simulation.Simulation results show that the proposed algorithm has fast convergence speed and is better than best response dynamics(BRD) algorithm and No-regret algorithm.

Clustering-based Demand Response for Intelligent Energy Management in 6G-enabled Smart Grids

Ran WANG, Jiang-tian NIE, Yang ZHANG, Kun ZHU

Computer Science. 2022, 49 (6): 44-54. doi:10.11896/jsjkx.220400002

Abstract

PDF(3070KB) ( 698 )

References | Related Articles | Metrics

As a typical industrial Internet of things (IIOT) service,demand response(DR) is becoming a promising enabler for intelligent energy management in 6G-enabled smart grid systems,to achieve quick response for supply-demand mismatches.How-ever,existing literatures try to adjust customers’ load profiles optimally,instead of electricity overhead,energy consumption patterns of residential appliances,customer satisfaction levels,and energy consumption habits.In this paper,a novel DR method is investigated by mixing the aforementioned factors,where the residential customer cluster is proposed to enhance the performance.Clustering approaches are leveraged to study the electricity consumption habits of various customers by extracting their features and characteristics from historical data.Based on the extracted information,the residential appliances can be scheduled effectively and flexibly.Moreover,we propose and study an efficient optimization framework to obtain the optimal scheduling solution by using clustering and deep learning methods.Extensive simulation experiments are conducted with real-world traces.Numerical results show that the proposed DR method and optimization framework outperform other baseline schemes in terms of the system overhead and peak-to-average ratio (PAR).The impact of various factors on the system utility is further analyzed,which provides useful insights on improving the efficiency of the DR strategy.With the achievement of efficient and intelligent energy management,the proposed method also promotes the realization of China’s carbon peaking and carbon neutrality goals.

Priority Based EV Charging Management Under Service Reservation in Smart Grid

ZHANG Jie, TANG Qiang, LIU Shuo-han, CAO Yue, ZHAO Wei, LIU Tao, XIE Shi-ming

Computer Science. 2022, 49 (6): 55-65. doi:10.11896/jsjkx.220200013

Abstract

PDF(3375KB) ( 624 )

References | Related Articles | Metrics

The introduction of electric vehicles(EVs) alleviates greenhouse gases emission.Its application has huge potential in the attempt to achieve green transportation today.However,the long charging time and charging congestion greatly affect the travel experience of EVs.To optimize EV charging,the charging station(CS) selection scheme(where to charge) and the charging scheduling strategy(when to charge) become the core of solving the problem of urban EV charging.In this paper,the preemptive charging scheduling strategy considering the charging priority(CP) is proposed.This strategy allows the preemptive charging of EVs with high urgency of charging(calculated from the charging demand and the remaining parking duration).Based on the CP charging scheduling strategy,a CS selection scheme that further combines reservation information is optimized.This scheme selects the CS with the shortest charging travel time(including one-time charging process) for EVs.Meanwhile,EVs are required to report their charging reservation information to accurately predict the congestion status of CSs,so as to efficiently allocate charging resources.The charging network is simulated through the urban traffic scene of Helsinki.The results show that the charging management scheme,CP scheduling strategy and reservation-based CS selection scheme proposed in this paper,can effectively shorten the average charging travel time of EVs and provide fully charging service for more EVs within a limited parking duration.

Tile Partition Optimized Omnidirectional Video Coding for 6G Network

YANG Tao-yu, XU Yuan-yuan, TAN Zeng-jie

Computer Science. 2022, 49 (6): 66-72. doi:10.11896/jsjkx.220400034

Abstract

PDF(2872KB) ( 684 )

References | Related Articles | Metrics

The rise of the 6G wireless communication provides a broader prospect for the development of virtual reality panoramic video.The tile-based panoramic video coding scheme can improve the viewing experience of 360° video under the same condition of network bandwidth.Tile partition affects the video transmission performance.Compared with small tiles,using large tiles can effectively improve coding efficiency,but it will cause more pixel overhead by transmitting a larger area to cover the viewport.The existing tile partition work is mainly designed for the rectangular viewport,but the projection of spherical video to a two-dimensional plane will stretch different areas of the viewport to varying degrees in practice.Deriving the pixel overhead associated with covering irregular viewport areas with rectangular tiles is more complicated.To address this challenge,a tile partitioning algorithm for the user's real viewport has been proposed in this paper.Firstly,the stretching distortion of the viewport caused by the projection format is analyzed,and the pixel overhead of the irregular viewport with different tile partition sizes is derived.Secondly,by trading off the pixel overhead and the coding efficiency of tiles with different granularity,an optimal tile partition scheme has been proposed for the panoramic video sequence.Finally,the proposed scheme is compared with the exhaustive search method for tile partition in the experiment,and the results show that the proposed algorithm can achieve almost the same transmission efficiency as the exhaustive search method with less computational complexity.

Study on Preprocessing Algorithm for Partition Reconnection of Unstructured-grid Based on Domestic Many-core Architecture

YE Yue-jin, LI Fang, CHEN De-xun, GUO Heng, CHEN Xin

Computer Science. 2022, 49 (6): 73-80. doi:10.11896/jsjkx.210900045

Abstract

PDF(3168KB) ( 602 )

References | Related Articles | Metrics

How to efficiently solve the discrete-memory-accessing problem of unstructed-grid is one of the hot-spot issues in the field of parallel algorithms and application in scientific and engineering computing.The distributed block reconnection optimization algorithm,which is designed on the basis of domestic Sunway heterogeneous many-core architecture,can maintain high computing performance when solving the problem of unstructured sparsity in applications.After deeply analyzing the on-chip communication mechanism of the many-core architecture,an efficient message grouping strategy is designed to improve the bandwidth utilization of on-chip array on the slave core.At the same time,a barrier-free data distribution algorithm is combined to give full play to the network perfor-mance of the domestic heterogeneous many-core architecture.Through the establishment of perfor-mance models and experimental analysis,the average memory bandwidth of the proposed algorithm can reach more than 70% of the theoretical value under different memory access situations.Compared with the serial algorithm on the master core,it has an ave-rage of 10 times and a maximum of 45 times performance acceleration.At the same time,the universal applicability of the algorithm is proved by application tests in different fields.

Architecture Design for Particle Transport Code Acceleration

FU Si-qing, LI Tie-jun, ZHANG Jian-min

Computer Science. 2022, 49 (6): 81-88. doi:10.11896/jsjkx.210600179

Abstract

PDF(2310KB) ( 602 )

References | Related Articles | Metrics

The stochastic simulation method of particle transport is usually used to solve the characteristic quantity of a large number of moving particles.Particle transport problems are widely found in the fields of medicine,astrophysics and nuclear phy-sics.The main challenge of current stochastic simulation methods for particle transport is the gap between the number of simulation samples supported by computers,the simulation timescale,and researchers’ needs to study practical problems.Since the development of processor performance has entered a new historical stage with the stagnation of process size progress,the integration of complex on-chip structures no longer meets the current requirements.For particle transport programs,this paper carries out a series of architecture design works.By analyzing and using the parallelism and access characteristics of the program,simplified kernel and reconfigurable cache are designed to speed up the program.Experiments show that compared to the traditional architecture composed of multiple out-of-order cores,this architecture can obtain more than 4.5x in performance per watt and 2.78x in performance per area,which lays a foundation for the further study of large-scale many-nucleus particle transport acce-lerator.

Survey on Multithreaded Data Race Detection Techniques

ZHAO Jing-wen, FU Yan, WU Yan-xia, CHEN Jun-wen, FENG Yun, DONG Ji-bin, LIU Jia-qi

Computer Science. 2022, 49 (6): 89-98. doi:10.11896/jsjkx.210700187

Abstract

PDF(1770KB) ( 820 )

References | Related Articles | Metrics

Nowadays the multi-core processors and threaded parallel programs are increasingly more used.However,the uncertainty of multi-threaded program leads to concurrency problems such as data race,atomicity violation,order violation and deadlock in the process of program running.Recent researches show that data race accounts for the largest proportion of concurrency bug,and most atomicity violation and order violation are caused by data race.This paper summarizes the related detection techniques in recent years.Firstly,the related concepts,causes,and the main ideas of data race detection are introduced.Then,the existing data race detection techniques in multi-threaded program are classified into three types:static analysis,dynamic analysis and hybrid detection techniques,and their characteristics are summarized comprehensively and compared in detail.Next,the limitations of exis-ting data race detection tools are discussed.Finally,future research directions and challenges in this field are discussed.

Parallel Optimization Method of Unstructured-grid Computing in CFD for DomesticHeterogeneous Many-core Architecture

CHEN Xin, LI Fang, DING Hai-xin, SUN Wei-ze, LIU Xin, CHEN De-xun, YE Yue-jin, HE Xiang

Computer Science. 2022, 49 (6): 99-107. doi:10.11896/jsjkx.210400157

Abstract

PDF(3943KB) ( 1075 )

References | Related Articles | Metrics

Sunway TaihuLight ranked first in the global supercomputer top 500 list 2016－2018 with a peak performance of 125.4 PFlops.Its computing power is mainly attributed to the domestic SW26010 many-core RISC processor.CFD unstructured-grid computing has always been a challenge for porting and optimizing in domestic many-core supercomputer,because of its complex topology,serious discrete memory access problems,and strongly correlated linear equation solution.In order to give fully play to the computing efficiency of domestic heterogeneous multi-core architecture,firstly,a data reconstruction model is proposed to improve the locality and parallelism of data,and the data structure is more suitable for the characteristics of multi-core architecture.Secondly,aiming at the discrete memory access problem caused by the disorder of unstructured-grid data storage,a discrete memory access optimization method based on prestorage of information relation is proposed,which transforms discrete memory access into continuous memory access.Finally,the pipeline parallelism mechanism in core array is introduced to realize many-core parallelism for solving linear equations with strong correlation.Experiments show that the overall performance of unstructured-grid computing in CFD is improved by more than 4 times,and is 1.2x faster than the general CPU.The computing cores scale to 624 000,and the parallelism efficiency is maintained at 64.5%.

GPU-based Parallel DILU Preconditioning Technique

WANG Jin, LIU Jiang

Computer Science. 2022, 49 (6): 108-118. doi:10.11896/jsjkx.210300259

Abstract

PDF(2690KB) ( 738 )

References | Related Articles | Metrics

Large sparse linear equations often appear in scientific computation and engineering.There are many iterative methods and preconditioning techniques for solving these linear equations.Diagonal-based incomplete LU (DILU) is a preconditioning technique similar to incomplete LU (ILU) factorization.DILU is applied in OpenFOAM,an open source computational fluid dynamics software,and is a very important preconditioning technique in OpenFOAM.DILU has not received extensive attention outside OpenFOAM,and there is no complete GPU-based implementation so far.This paper compares DILU preconditioned BiCGStab with ILU preconditioned BiCGStab,and the time elapses in preconditioner constructions.The numeric experiments suggest that DILU may be more efficient and stable than ILU.As for GPU-based parallel implementations,this paper discusses two parallel schemes,that are level-set scheme and synchronization-free scheme,and gives related algorithms and some codes under these two parallel schemes.It compares the performances of DILU preconditioning technique under two parallel schemes.The numeric results show that each scheme has its own advantages and disadvantages in different equations,and we can select one according to their performances in practice.This paper compares the performance of DILU preconditioning on GPU and CPU,and the results show that GPU is more competitive.The applications that have performance bottlenecks on linear systems solutions can be improved by moving to GPU platforms.

Study on Temporal Influence Maximization Driven by User Behavior

WEI Peng, MA Yu-liang, YUAN Ye, WU An-biao

Computer Science. 2022, 49 (6): 119-126. doi:10.11896/jsjkx.210700145

Abstract

PDF(2516KB) ( 509 )

References | Related Articles | Metrics

Influence maximization(IM) aims to find a group of users in a social network,through whom information can spread most widely in the network.Existing studies mainly focus on the IM problem in static networks.However,social networks are constantly evolving in real life,and propagation models(such as independent cascading model and linear threshold model) based on static networks are not suitable for the information propagation process in evolving networks.Meanwhile,the existing researches ignore the influence of user behavior on information propagation.Therefore,to tackle this problem,this paper proposes a behavior driven independent cascade(BDIC) propagation model,which can effectively describe the information propagation process in the evolving social networks.Based on this model,a user behavior-driven IM algorithm is proposed.It mainly includes three steps.Firstly,the process of message transmission is modeled to calculate the probability of information transmission in evolving social networks.Then,a user behavior-driven reverse influence sampling algorithm is proposed,which can effectively query the most influential user with a specific time.Finally,a seed query algorithm under different time(time series) is designed,which can effectively reflect the dynamic change characteristics of seed nodes in evolving social networks.To evaluate the effectiveness of the proposed algorithm,a similarity comparison method between seed nodes and the affected nodes is designed.Experiments on real datasets verify the efficiency and scalability of the proposed approaches.The results also demonstrate that the BDIC model can effectively reflect the information propagation process in evolving social networks.

Tri-training Algorithm Based on DECORATE Ensemble Learning and Credibility Assessment

WANG Yu-fei, CHEN Wen

Computer Science. 2022, 49 (6): 127-133. doi:10.11896/jsjkx.211100043

Abstract

PDF(2041KB) ( 604 )

References | Related Articles | Metrics

Tri-training is a disagreement-based semi-supervised learning algorithm,in which both semi-supervised learning and ensemble learning mechanisms are simultaneously applied.It can improve the model performance by effectively leveraging some labeled samples along with a large amount of unlabeled ones through collaborations and iterations among basic classifiers.How-ever,when the labeled sample size is insufficient,the initial classifiers generated by Tri-training are not sufficiently trained.Furthermore,mislabeled noisy data might be generated during the collaborative labeling process among the classifiers.Aiming at these problems,a collaborative learning algorithm is proposed,which combines DECORATE ensemble learning,diversity mea-sure and credibility assessment.In our method,to improve the generalization performance,multiple preference classifiers are generated based on DECORATE with differentiated artificial data and labels,and the diversities of classifiers are measured and selected by Jensen-Shannon divergence to maxmize the diversity of the classifiers.At the same time,the credibility of the pseudo labeled samples is assessed during the iterations by a label propagation algorithm to reduce the noisy data.The results of classification experiment on UCI data sets demonstrate that the proposed algorithm achieves higher accuracy and F1-score than Tri-trai-ning algorithm and its improved versions.

Adaptive Weight Based Broad Learning Algorithm for Cascaded Enhanced Nodes

CAI Xin-yu, FENG Xiang, YU Hui-qun

Computer Science. 2022, 49 (6): 134-141. doi:10.11896/jsjkx.210500119

Abstract

PDF(3271KB) ( 709 )

References | Related Articles | Metrics

In the era of intelligence,continuous autonomous learning and optimization need to be carried out on the big data platform,and the first step of continuous autonomous learning is data enhancement.This paper proposes a broad learning method based on cascaded enhancement nodes,which provides a new data enhancement method for continuous autonomous learning on big data platform,and makes it possible for subsequent evolutionary optimization on the basis of learning architecture.Classical broad learning is a typical feedforward neural network,which is not suitable for modeling dynamic time series.In this paper,the feedback structure is introduced into the traditional broad learning system,which makes the enhancement nodes have memory and retains part of the historical information.In feature extraction,phase space reconstruction is used to extract more essential features of the data.At the same time,a weight factor is introduced to assign different weights to each sample according to its contribution to model during training,so as to eliminate the interference of noise and outliers to the learning process and improve the robustness of the algorithm.Experimental results show that the proposed algorithm is effective.

Personalized News Recommendation Algorithm with Enhanced List Information and User Interests

PU Qian-qian, LEI Hang, LI Zhen-hao, LI Xiao-yu

Computer Science. 2022, 49 (6): 142-148. doi:10.11896/jsjkx.210400173

Abstract

PDF(2449KB) ( 589 )

References | Related Articles | Metrics

With the continuous expansion of data and information,the point-to-point recommendation model,as a commonly used recommendation algorithm in deep learning,can deal with the problem of overloaded information to some extent.However,it predicts the recommendation score only by a single user and an isolated news,without using of the interactive information among rele-vant lists of news.To improve the quality of personalized recommendation,it is urgent for current news recommendation platforms to figure out how to accurately and comprehensively represent users and news by taking full advantage of users’ browsing history,semantic meaning of news as well as list information.In view of this,this paper puts forward a personalized news recommendation algorithm with improved list information and user interest.Based on the historically browsed news sequence of the user and news data,the point-to-point recommendation model is trained for representation construction to realize the tailored information extraction catering to the users’ interest,and the list information is enhanced by processing the characteristics of the user and news lists through the attention network,thus realizing the direct recommendation ranking of the lists as a whole.Experimental results show that this personalized recommendation algorithm with enhanced list information and user attraction can model global the comprehensive list information,presenting a significantly improved effect compared with cutting-edge news re-commendation algorithms at present.

Study on Intelligent Recommendation Method of Dueling Network Reinforcement Learning Based on Regret Exploration

HONG Zhi-li, LAI Jun, CAO Lei, CHEN Xi-liang, XU Zhi-xiong

Computer Science. 2022, 49 (6): 149-157. doi:10.11896/jsjkx.210600226

Abstract

PDF(3034KB) ( 615 )

References | Related Articles | Metrics

In recent years,the application of deep reinforcement learning in recommendation system has attracted much attention.Based on the existing research,this paper proposes a new recommendation model RP-Dueling,which is based on the deep reinforcement learning Dueling-DQN algorithm,and adds the regret exploration mechanism to make the algorithm adaptively and dynamically adjust the proportion of “exploration-utilization” according to the training degree.The algorithm can capture users’ dynamic interest and fully explore the action space in the recommendation system with large-scale state space.By testing the proposed algorithm model on multiple data sets,the optimal average results of MAE and RMSE are 0.16 and 0.43 respectively,which are 0.48 and 0.56 higher than the current optimal research results.Experimental results show that the proposed model is superior to the existing traditional recommendation model and recommendation model based on deep reinforcement learning.

Hybrid Recommender System Based on Attention Mechanisms and Gating Network

GUO Liang, YANG Xing-yao, YU Jiong, HAN Chen, HUANG Zhong-hao

Computer Science. 2022, 49 (6): 158-164. doi:10.11896/jsjkx.210500013

Abstract

PDF(2657KB) ( 901 )

References | Related Articles | Metrics

Combining user reviews with user ratings to improve the performance of recommender system is the current mainstream research direction of recommender system.However,when user review data is sparse,the performance of most existing recommender systems will degrade to a certain extent.To solve this problem,this paper proposes a hybrid recommendation system (AMGNRS),which combines attention mechanism and gating networking based recommendation system.It use auxiliary comments generated by like-minded users to alleviate the sparsity of user comments.Firstly,a variety of mixed attention mechanism are combined to impove the feature extracting efficiency of user comments and grading.Then features are extracted by adaptive fusion of gated network,and features most relevant to user preference are selected.Finally,the higher order linear interaction of the neural factorization machine is used to derive the score prediction.By comparing the model with the current model with excellent performance on three real data sets,the results show that the problem of data sparsity is significantly alleviated and the effectiveness of the model is verified.

Graph Neural Network Recommendation Model Integrating User Preferences

XIONG Zhong-min, SHU Gui-wen, GUO Huai-yu

Computer Science. 2022, 49 (6): 165-171. doi:10.11896/jsjkx.210400276

Abstract

PDF(2137KB) ( 840 )

References | Related Articles | Metrics

Aiming at the problem that knowledge graph-driven graph neural network recommendation algorithm cannot learn the user and item representations at the same time,a graph neural network recommendation model that integrates user preferences is proposed.The model learns user and item representations from user’s perspective and entity’s perspective respectively.Firstly,the user’s perspective spreads user preferences in the knowledge graph based on user historical interaction records and enhances user representation.Secondly,the entity perspective gathers neighbor information of candidate entities through graph convolu-tional network to enrich the representation of the entity.At the same time,a hybrid layer is designed to capture high-level connectivity and hybrid hierarchical information from both the width and depth aspects to enhance the item representation.The enhanced user representation vector and item representation vector are input to the prediction function to predict the interaction probability.Finally,the fixed-size sampling method and phased training strategy are used to optimize the model.The click-through rate prediction experiment is conducted on the MovieLens-1M data set,and the results show that,compared with the benchmark methods RippleNet and KGCN,its AUC increases by 1.7% and 2.3% respectively.

Optimal Scale Selection in Random Multi-scale Ordered Decision Systems

FANG Lian-hua, LIN Yu-mei, WU Wei-zhi

Computer Science. 2022, 49 (6): 172-179. doi:10.11896/jsjkx.220200067

Abstract

PDF(1586KB) ( 421 )

References | Related Articles | Metrics

Aiming at the knowledge acquisition problem of multi-scale ordered information system obtained from random experiments,concepts of random multi-scale ordered information systems and dominance-equivalence-relations-based random multi-scale ordered decision systems are first introduced.Information granules in random multi-scale ordered information systems as well as lower and upper approximations of sets with respect to dominance relations induced by conditional attribute set under different scales are then described.Their relationships are also clarified.Finally,concepts of several types of optimal scales in random multi-scale ordered information systems and dominance-equivalence-relations-based random multi-scale ordered decision systems are defined.It is proved that belief and plausibility functions in the Dempster-Shafer theory of evidence can be used to characterize some optimal scales in random multi-scale ordered information systems and dominance-equivalence-relations-based random multi-scale ordered decision systems,respectively.

Stylized Image Captioning Model Based on Disentangle-Retrieve-Generate

CHEN Zhang-hui, XIONG Yun

Computer Science. 2022, 49 (6): 180-186. doi:10.11896/jsjkx.211100129

Abstract

PDF(2360KB) ( 708 )

References | Related Articles | Metrics

Image captioning aims to generate a description text for the input image to accurately describe the image content.The stylized image captioning goes a step further on the basis of image captioning and introduces the consideration of language style.It also needs appropriately express the specific language style,which makes the generated text more diverse.In order to better incorporate style factors to the description text,a stylized image captioning model based on disentangle-retrieve-generate framework is proposed.The model first splits the sentences in the stylized corpus into content and style parts,and constructs a content-style memory module,then retrieves appropriate style from the memory module according to the factual caption of the image.Finally,the factual caption and retrieved style part are input into the language model for stylized caption generation.Experimental results on real datasets show that,compared to existing methods,the proposed model has better performance in various evaluation me-trics,and can accurately describe the image content while expressing a specific style.

Speech Enhancement Based on Time-Frequency Domain GAN

YIN Wen-bing, GAO Ge, ZENG Bang, WANG Xiao, CHEN Yi

Computer Science. 2022, 49 (6): 187-192. doi:10.11896/jsjkx.210500114

Abstract

PDF(2613KB) ( 722 )

References | Related Articles | Metrics

The traditional speech enhancement algorithm based on generative adversarial networks (SEGAN) enhances speech in the time domain,and completely ignores the distribution of speech samples in frequency domain.Under the condition of low signal-to-noise ratio,the speech signal will be submerged in noise,and the time-domain distribution information of noisy speech is difficult to capture.Therefore,the enhancement performance of SEGAN will drop sharply,and the speech quality and speech intelligibility of its enhanced speech are very low.To solve this problem,this paper proposes a speech enhancement algorithm (time-frequency domain SEGAN,TFSEGAN) based on time-frequency domain generation confrontation network.TFSEGAN adopts the model structure of the time-frequency domain dual discriminator,and a time-frequency L1 loss function.The input of time domain discriminator is time domain feature of the speech sample,and the input of frequency domain discriminator is frequency domain feature of the speech sample.In the training process,time-domain discriminator uses the time-domain distribution information of speech sample as the criterion,and frequency-domain discriminator uses the frequency-domain distribution information of the speech sample as the criterion.Under the action of two discriminators,the generator of TFSEGAN could simulta-neously learn the distribution rules and information of speech samples in time domain and frequency domain.Experiments prove that,compared with SEGAN,the speech quality and intelligibility of TFSEGAN improve by about 17.45% and 11.75% respectively at low signal-to-noise ratio.

Remote Sensing Change Detection Based on Feature Fusion and Attention Network

LAN Ling-xiang, CHI Ming-min

Computer Science. 2022, 49 (6): 193-198. doi:10.11896/jsjkx.210500058

Abstract

PDF(2372KB) ( 658 )

References | Related Articles | Metrics

Change detection is one of the essential tasks in remote sensing,which is usually regarded as a pixel-level classification problem.In recent years,deep neural networks have also been widely used in the change detection task due to their powerful hierarchical representation of bi-temporal images.A feature fusion and attention network (FFAN) is proposed based on neural encoder-fusion-decoder framework.It integrates features generated by encoder with the bi-temporal difference feature enhanced by attention mechanism,to better capture the bi-temporal change information.In particular,bi-temporal features enhanced by attention mechanism can significantly enhance the propagation of change information in the intermediate layers of deep networks,which adaptively recalibrates the change activation in FFAN by explicitly modeling the interdependence of bi-temporal inputs.Experiments conducted on open-source dataset demonstrate that,compared with existing methods,FFAN obtains better performance.

Low-light Image Enhancement Based on Retinex Theory by Convolutional Neural Network

ZHAO Zheng-peng, LI Jun-gang, PU Yuan-yuan

Computer Science. 2022, 49 (6): 199-209. doi:10.11896/jsjkx.210400092

Abstract

PDF(7271KB) ( 708 )

References | Related Articles | Metrics

In the course of decomposing and enhancing the low-light images with Retinex model,it needs to manually adjust the parameters continuously to reach the optimal solution,which will reduce the efficiency of the entire process.In addition,existing low-light image enhancement methods based on Retinex fail to take both reflectance and illumination into account when perfor-ming image enhancement,and there are problems such as too much noise in the reflectance of low-light image,low brightness and not enough prominent details in the illumination.Aiming to solve these problems,a data-driven deep network is proposed to learn the decomposition and the enhancement of the low-light images,and the model parameters are learned through the end-to-end network training.The network firstly decomposes the low-light images into the reflectance and the illumination.Aiming at the problem of high noise in the reflectance,an improved denoising convolutional neural network model NDnCNN is used for denoising,and aiming at the problems of low brightness and not enough prominent details in the illumination,we introduce the convolutional block attention model CBAM to enhance the details and guide the network to modify the illumination.Finally,the denoised reflectance and the modified illumination are used for image reconstruction.Experimental results show that the enhanced low-light image is more photo-realistic with increased brightness,prominent details,rich information and low image distortion.

Study on Multi-label Image Classification Based on Sample Distribution Loss

ZHU Xu-dong, XIONG Yun

Computer Science. 2022, 49 (6): 210-216. doi:10.11896/jsjkx.210300267

Abstract

PDF(2289KB) ( 740 )

References | Related Articles | Metrics

Different from the data distribution in general image classification scenarios,in the scenario of multi label image classification,the sample number distribution among different label categories is unbalanced,and a small number of head categories often account for the majority of sample size.However,due to the correlation between multiple labels,and the distribution of diffi-cult samples under multiple labels is also related to the data distribution and category distribution,the re-sampling and other methods for solving the data imbalance in the single label problem cannot be effectively applied in the multi label scenario.This paper proposes a classification method based on the loss of sample distribution in multi label image scene and deep learning.Firs-tly,the unbalanced distribution of multi label data is set with category correlation,and the loss is re-used,and the dynamic lear-ning method is used to prevent the excessive alienation of distribution.Then,the asymmetric sample learning loss is designed,and different learning abilities for positive and negative samples and difficult samples are set.At the same time,the information loss is reduced by softening the sample learning weight.Experiments on related data sets show that the algorithm has achieved good results in solving the sample learning problem in the scene of uneven distribution of multi-label data.

Digital Mural Inpainting Method Based on Feature Perception

XU Hui, KANG Jin-meng, ZHANG Jia-wan

Computer Science. 2022, 49 (6): 217-223. doi:10.11896/jsjkx.210500105

Abstract

PDF(3787KB) ( 709 )

References | Related Articles | Metrics

There are irregular damaged areas caused by various diseases of grottoes in Dunhuang murals,digital restoration is used to restore the image of the Dunhuang grotto murals,which will not cause damage to the original murals,but also get a better repair effect.Because of the large missing area in the mural mending,it cannot be realized by local non-semantic repair methods.Aiming at the restoration of the defective area of Dunhuang grotto murals,this paper designs an image repair method based on the generation of confrontation network,and uses semantically reasonable content to render the pixels in the missing area to realize the reconstruction of non-contact mural scenes,improve the efficiency of mural virtual restoration and the accuracy of restoration.The algorithm introduces a perceptual-loss function on the basis of generating an adversarial neural network,adds a three-layer convolutional layer to the generation model to collect image features of damaged areas,uses the perceptual loss to improve the model’s ability to repair high-frequency texture details,and uses extended convolution to extract range features,so as to stimulate the generative model to generate higher quality image results.Compared with three excellent methods on the Dunhuang grotto mural data set,and the repair results show that the PSNR score of the proposed algorithm on the test data set increases by 1.79%,and the SSIM score increases by 7.7%.The proposed repair model improves the repair accuracy of damaged murals and makes the repair results more accurate.

Multi-branch RA Capsule Network and Its Application in Image Classification

WU Lin, SUN Jing-yu

Computer Science. 2022, 49 (6): 224-230. doi:10.11896/jsjkx.210400087

Abstract

PDF(3123KB) ( 403 )

References | Related Articles | Metrics

Capsule Network is a new type of deep neural network that uses vectors to express information of image feature and overcomes two major problems of convolutional neural networks by introducing dynamic routing algorithms.First,convolutional neural networks cannot learn and express the part-whole relationship of images.Second,pooling operations lead to serious loss of image feature information.However,CapsNet needs to learn all the features of the image,and when the image background is complex,it has the problems of insufficient information of extracted image features,large number of training parameters and low training efficiency.To this end,firstly,a lightweight image feature extractor RA module is designed to extract image feature information faster and more completely.Secondly,two different depths of lightweight branches are designed to improve the training efficiency of the network.Finally,a new compression function hc-squash is designed to ensure that the network can acquire more useful information,and a multi-branch RA (Resnet Attention) capsule network is proposed.Through the application in the four image classification datasets of MNIST,Fashion-MNIST,affNIST and CIFAR-10,it is confirmed that the multi-branch RA capsule network outperforms CapsNet and MLCN in several performance metrics,and an improvement scheme is designed for the proposed network to achieve optimised classification performance.

Multi-threshold Segmentation for Color Image Based on Pyramid Evolution Strategy

XU Ru-li, HUANG Zhang-can, XIE Qin-qin, LI Hua-feng, ZHAN Hang

Computer Science. 2022, 49 (6): 231-237. doi:10.11896/jsjkx.210300096

Abstract

PDF(3444KB) ( 439 )

References | Related Articles | Metrics

In view of the fact that traditional intelligent optimization algorithms for multi-threshold segmentation of color images fall to consider the competition and cooperation between populations,which results in local optimization problems that affect the segmentation effect.In order to improve the segmentation effect,an improved pyramid evolution strategy (IPES) is proposed.The algorithm designs an adaptive search operator suitable for the multi-threshold segmentation problem of color images;expands the search space at all levels,improves the optimization ability of the algorithm;takes Otsu as the optimization goal and uses the competition and cooperation relationship between populations to solve the local optimization problem,thereby improving the accuracy of the solution and the effect of segmentation.The performance of IPES is tested on existing standard test images and compared with other eight algorithms.Experimental results show that the peak signal-to-noise ratio of the image segmented by IPES algorithm is between 28~35 dB,which is at least 10 dB higher than that of the improved tree-seed algorithm and traditional particle swarm algorithm and differential evolution algorithm;the structural similarity is between 89%~97%,increased by at least 3%.The image quality after segmentation is better and the structural similarity is higher.Therefore,the algorithm has good perfor-mance in solving multi-threshold segmentation problem of color images.

Small Object Detection in 3D Urban Scenes

CHEN Jia-zhou, ZHAO Yi-bo, XU Yang-hui, MA Ji, JIN Ling-feng, QIN Xu-jia

Computer Science. 2022, 49 (6): 238-244. doi:10.11896/jsjkx.210400174

Abstract

PDF(3491KB) ( 625 )

References | Related Articles | Metrics

3D object detection is the core of semantic analysis in 3D urban scenes,but the existing object detection methods mainly focus on large objects such as buildings and roads,while the detection accuracy of these methods for small objects such as street lamps and manhole covers is low.For this sake,a multi-view small object detection method for 3D urban scenes is proposed.It combines the oblique photogrammetry and 3D object localization,to improve the detection accuracy of small objects.Firstly,small objects are detected in the UAV images using a deep neural network.Then,detection results are back projected onto the three-dimensional urban model.Finally,the 3D detection results are obtained by clustering these 3D objects obtained by back projection.Experimental results show that the proposed method can automatically detect small objects such as manhole covers and windows on the large-scale 3D urban model reconstructedby oblique photogrammetry,it is free of spatial occlusion,and has high accuracy and stability compared with object detection on orthophoto maps.

Compression Algorithm of Face Recognition Model Based on Unlabeled Knowledge Distillation

CHENG Xiang-ming, DENG Chun-hua

Computer Science. 2022, 49 (6): 245-253. doi:10.11896/jsjkx.210400023

Abstract

PDF(3390KB) ( 700 )

References | Related Articles | Metrics

When transplanting face recognition technology to mobile devices,it often needs to be processed by accelerated algorithms such as model compression.Knowledge distillation is a model compression method that has a wide range of practical applications and is easy to train.Existing knowledge distillation algorithms require a large amount of tagged face data,which may involve security issues such as identity privacy leakage.At the same time,the cost of large-scale collection of tagged face data is relatively high,while the massive amount of unlabeled face data that can be collected or generated cannot be used.In order to solve the above problems,this paper analyzes the characteristics of knowledge distillation in face recognition tasks,and proposes an indirect supervised training method of unlabeled knowledge distillation.This method can utilize massive amounts of unlabeled face data,thereby avoiding security risks such as privacy leakage.However,the data distribution of the unlabeled face data set is unpredictable,and there is the problem of uneven data distribution,which limits the performance of the indirect supervision algorithm.This research further proposes a data enhancement method for face content replacement,which balances the distribution of face data by replacing part of the content of the face,and at the same time enhances the diversity of face data.Sufficient experimental results show that when the face recognition model is greatly compressed,the performance of the algorithm in this research reaches an advanced level,and surpasses the large-scale network on the LFW data set.

Aerial Violence Recognition Based on Spatial-Temporal Graph Convolutional Networks and Attention Model

SHAO Yan-hua, LI Wen-feng, ZHANG Xiao-qiang, CHU Hong-yu, RAO Yun-bo, CHEN Lu

Computer Science. 2022, 49 (6): 254-261. doi:10.11896/jsjkx.210400272

Abstract

PDF(3756KB) ( 777 )

References | Related Articles | Metrics

The violence in public areas occurs frequently and video surveillance is of great significance for maintaining public safety.Compared with fixed cameras,unmanned aerial vehicles (UAVs) have surveillance mobility.However,in aerial images,the rapid movement of UAVs as well as the change of posture and height cause the problem of motion blur and large-scale change of target.To solve this problem,an attention spatial-temporal convolutional network (AST-GCN) combining attention mechanism is designed to realize the identification of violent behavior in aerial video.The proposed method is divided into two steps:the key frame detection network completes the initial positioning,and the AST-GCN network completes the behavior identification through the sequence features.Firstly,aiming at video violence localization,a key frame cascade detection network is designed to realize violence key frame detection based on human posture estimation,and preliminarily judge the occurrence time of violence.Secondly,the skeleton information of multiple frames around key frames is extracted from the video sequence,and the skeleton data is pre-processed,including normalization,screening and completion,so as to improve the robustness of different scenes and the partial missing of key nodes.And the skeleton temporal-spatial representation matrix is constructed according to the extracted skeleton information.Finally,AST-GCN network analyzes and identifies multiple frames of human skeleton information,to integrate attention module,improve feature expression ability,and complete the recognition of violent behavior.The method is validated on self-built aerial violence data set,and experimental results show that the AST-GCN can realize the recognition of aerial scene violence,and the recognition accuracy is 86.6%.The proposed method has important engineering value and scientific signifi-cance for the realization of aerial video surveillance and human pose understanding applications.

Fast Structural Texture Image Synthesis Algorithm Based on Seam ConsistencyCriterion

JIN Li-zhen, LI Qing-zhong

Computer Science. 2022, 49 (6): 262-268. doi:10.11896/jsjkx.210400039

Abstract

PDF(7081KB) ( 547 )

References | Related Articles | Metrics

Aiming at the problems of patch-based synthesis algorithm of structured texture images,such as discontinuity of structure,distortion of boundary,seam misalignment,and low synthesis speed,a new fast non-overlapping synthesis algorithm of texture images is proposed based on the consistency criterion of double-seam lines,thereby effectively improving the synthesis quality and speed of structured texture images.Firstly,the seamline consistency criterion considering hue,saturation,brightness and edge characteristics simultaneously is established in HSI color space that is more consistent with human visual characteristic.Then,a sub-block search strategy and a new non-overlapping splicing algorithm based on the consistency criterion of double-seam line are proposed and implemented.The experiment results show that the proposed algorithm can significantly improve the synthesis quality and speed of structured texture images in comparison with the traditional algorithms.

Multi-algorithm Fusion Behavior Classification Method for Body Bone Information Reconstruction

ZHAO Xiao-hu, YE Sheng, LI Xiao

Computer Science. 2022, 49 (6): 269-275. doi:10.11896/jsjkx.210500070

Abstract

PDF(4024KB) ( 596 )

References | Related Articles | Metrics

Aiming at the poor measurement effect of behavior monitoring in real life,a new method of extracting human behavior features is proposed.It not only considers the body point information,but also integrates the environmental attribute information of image.Considering that a large number of existing experiments use a variety of complex algorithms for experimental classification on the basis of human body features extraction,it does not take into account the irrationality of only using the body features for algorithm evaluation.Therefore,an image information reconstruction method based on body features is proposed in this paper,which combines the image convolution network of body features,attention mechanism and image recognition method to realize human behavior recognition.The body point information is extracted by Openpose,and then the body points are classified by graph convolution and attention.On the basis of the first classification,the body point expansion coefficient is added to segment the images so as to realize second accurate classification.Finally,the evaluation accuracy on the HMDB51 dataset improves by 5.6%,and it has a big advantage in the actual test.This shows that the method is not only more accurate,but also has more practical application value.

Application of Machine Learning in Financial Asset Pricing:A Review

XU Jie, ZHU Yu-kun, XING Chun-xiao

Computer Science. 2022, 49 (6): 276-286. doi:10.11896/jsjkx.210900127

Abstract

PDF(2098KB) ( 2599 )

References | Related Articles | Metrics

The key problem of financial asset allocation is asset price.Asset pricing is the core content of modern finance,which indicates that asset pricing law has always been one of the hot topics of financial research.This paper reviews the methods used by machine learning in the field of asset pricing and research progresses,classifies machine learning asset pricing method into machine learning method based on the characteristics processing and deep learning method based on end-to-end processing,compares the differences between different algorithms in principle and application scenarios,points out the applicability and limitations of the two kinds of machine learning methods,prospects the research direction on machine learning asset pricing in the future.

Survey on Online Adversarial Planning for Real-time Strategy Game

LUO Jun-ren, ZHANG Wan-peng, LU Li-na, CHEN Jing

Computer Science. 2022, 49 (6): 287-296. doi:10.11896/jsjkx.210600168

Abstract

PDF(2858KB) ( 782 )

References | Related Articles | Metrics

Real-time strategy game online adversarial planning is a challenging problem in the field of multi-agent learning.In the process of game confrontation,in the face of an uncertain threat environment and non-stationary opponents,the agent needs to reason about the opponent’s actions within a limited time according to the game situation,make your own action plan quickly and perform adversarial planning in the huge state space and action space.The real-time strategy game platform is an ideal testbed for studying online adversarial planning problems.This paper firstly uses a typical real-time strategy game model to elicit the real-time strategy game confrontation problems,and classifies them into three levels and two operation control methods,and sorts out the five challenges faced from five sub-directions.Secondly,the current online adversarial planning methods are comprehensively reviewed and analyzed from three perspectives of tactical adversarial planning,strategic adversarial planning and mixed adversarial planning.Finally,the key issues that need to be studied in the future are pointed out from three key aspects:opponent and player modeling,human-machine collaborative online ad hoc planning,and learning-based planning.

Hidden Preference-based Multi-objective Evolutionary Algorithm Based on Chebyshev Distance

SUN Gang, WU Jiang-jiang, CHEN Hao, LI Jun, XU Shi-yuan

Computer Science. 2022, 49 (6): 297-304. doi:10.11896/jsjkx.210500095

Abstract

PDF(3075KB) ( 1070 )

References | Related Articles | Metrics

As an important branch of multi-objective optimization,preference-based multi-objective evolutionary algorithms have been widely used in scientific researches and engineering practices,which have important research significance.In order to obtain the extreme solutions and the knee solution with the most compromised performance over each optimization objective in multi-objective optimization problems,a definition of knee solution based on Chebyshev distance and its geometric interpretation is presented.Based on the definition,a multi-objective evolutionary algorithm HP-NSGA-II aiming to search for the extreme solutions and the knee solution is proposed.The regional dynamic updating strategy of the proposed algorithm updates the target regions dynamically in each iteration,and finally converges to the target regions.The strategy of maintaining the balance between regions ensures the balance of the number of individuals in each region,so that the individuals could be distributed evenly in each region.Based on widely used test functions,sufficient experimental verification is carried out,and the experimental results indicate that HP-NSGA-II algorithm can achieve better performance in terms of convergence,regional balance and regional controllability in two-dimensional and three-dimensional test problems,and can accurately obtain the extreme solutions and knee solution.

Machine Translation Method Integrating New Energy Terminology Knowledge

DONG Zhen-heng, REN Wei-ping, YOU Xin-dong, LYU Xue-qiang

Computer Science. 2022, 49 (6): 305-312. doi:10.11896/jsjkx.210500117

Abstract

PDF(3256KB) ( 630 )

References | Related Articles | Metrics

In domain machine translation,whether domain terms can be translated correctly plays a decisive role in translation quality.It is of practical significance to effectively integrate domain terms into neural machine translation model and improve the translation quality of domain terms.This paper proposes a method to integrate the term information in the field of new energy into neural machine translation as a priori knowledge.Taking the term dictionary constructed by the bilingual term knowledge base in the field of new energy as the medium,this paper puts forward and compares two different ways of knowledge integration:1)term replacement,that is,replacing the source term with the target term at the source language end;2)term addition refers to the splicing of source side terms and target side terms at the source language side,the identifier as special external knowledge is used to identify the beginning and end of the target term at both the source language end and the target language end.Experiments are carried out based on the Chinese and English bilingual alignment corpus in the field of new energy and the constructed Chinese and English alignment corpus.The results show that on the test set,the Bleu value of the proposed method is 6.38 and 6.55 higher than that of the baseline experiment respectively,which proves that the proposed method can effectively integrate the domain term knowledge into the translation model and improve the translation quality of domain terms.

Automatic Summarization Model Combining BERT Word Embedding Representation and Topic Information Enhancement

GUO Yu-xin, CHEN Xiu-hong

Computer Science. 2022, 49 (6): 313-318. doi:10.11896/jsjkx.210400101

Abstract

PDF(1784KB) ( 682 )

References | Related Articles | Metrics

Automatic text summarization can help people to filter and identify information quickly,grasp the key content of news,and alleviate the problem of information overload.The mainstream abstractive summarization model is mainly based on the encoder-decoder architecture.In view of the fact that the decoder does not fully consider the text topic information when predicting the target word,and the traditional Word2Vec static word vector cannot solve the polysemy problem,an automatic summarization model for Chinese short news is proposed,which integrates the BERT word embedding representation and topic information enhancing.The encoder combines unsupervised algorithm to obtain text topic information and integrates it into the attention mechanism to improve the decoding effect of the model.At the decoder side,the BERT sentence vector extracted from the BERT pre-trained language model is used as the supplementary feature to obtain more semantic information.Meanwhile,pointer mechanism is introduced to solve the problem of out of vocabulary,and coverage mechanism is used to suppress repetition effectively.Finally,in the training process,reinforcement learning method is adopted to optimize the model for non-differentiable index ROUGE to avoid exposing bias.Experimental results on two datasets of Chinese short news summarization show that the proposed model can significantly improve the ROUGE evaluation index,effectively integrate text topic information,and generate fluent and concise summaries.

Chinese Entity Relations Classification Based on BERT-GRU-ATT

ZHAO Dan-dan, HUANG De-gen, MENG Jia-na, DONG Yu, ZHANG Pan

Computer Science. 2022, 49 (6): 319-325. doi:10.11896/jsjkx.210600123

Abstract

PDF(1897KB) ( 1105 )

References | Related Articles | Metrics

As the basic task of natural language processing,entity relations classification plays a critical role in tasks such as knowledge graphs,intelligent question answering,semantic web construction and so on.This paper constructs the BERT-GRU-ATT model to classify Chinese entity relations.In order to eliminate the influence of Chinese word segmentation ambiguity on entity relations classification,the pre-training model BERT(bi-directional encoder representations from transformers) is introduced as the embedding layer to better obtain the context information of Chinese characters.Then gate recurrent unit(GRU) is used to capture the long-distance dependence of entities in sentences and self-attention mechanism(ATT) is used to strengthen the weight of characters that contribute significantly to relations classification,so as to obtain better results of entity relations classification.In order to enlarge the Chinese entity relations classification corpus,we translate the SemEval2010_Task8 English entity relations evaluation corpus into Chinese.The model achieves an F1 value of 75.46% on this translation corpus,which shows the effectiveness of the proposed model.In addition,the model achieves an F1 of 80.55%on the SemEval2010-Task8 English dataset,which proves that the model has certain generalization ability to English corpus.

Text Classification Based on Attention Gated Graph Neural Network

DENG Zhao-yang, ZHONG Guo-qiang, WANG Dong

Computer Science. 2022, 49 (6): 326-334. doi:10.11896/jsjkx.210400218

Abstract

PDF(2248KB) ( 574 )

References | Related Articles | Metrics

To address the problem that the existing text classification work usually ignores the semantic interaction between words when generating text representation,this paper proposes a novel text classification model based on attention gated graph neural network.It makes effective use of the semantic features of words and improves the accuracy of text classification based on the adequate semantic interaction.Firstly,each input text is converted to a single graph-structured data and the semantic features of word nodes are extracted.Secondly,attention gated graph neural network is used to interact and update the semantic features of word nodes.In addition,the attention-based text pooling module is used to extract the word nodes with discriminative semantic features to construct text graph representation.Finally,effective text classification is implemented based on the text graph representation.Experimental results show that the proposed method achieves an accuracy of 70.83%,98.18%,94.72% and 80.03% on Ohsumed,R8,R52 and MR datasets,respectively,and outperforms existing methods.

Off-policy Maximum Entropy Deep Reinforcement Learning Algorithm Based on RandomlyWeighted Triple Q -Learning

FAN Jing-yu, LIU Quan

Computer Science. 2022, 49 (6): 335-341. doi:10.11896/jsjkx.210300081

Abstract

PDF(2414KB) ( 618 )

References | Related Articles | Metrics

Reinforcement learning is an important branch of machine learning.With the development of deep learning,deep reinforcement learning research has gradually developed into the focus of reinforcement learning research.Model-free off-policy deep reinforcement learning algorithms for continuous control attract everyone’s attention because of their strong practicality.Like Q-learning,algorithms based on actor-critic suffer from the problem of overestimations.To a certain extent,clipped double Q-lear-ning method solves the effect of the overestimation in actor-critic algorithms,but it also introduces underestimation to the lear-ning process.In order to further solve the problems of overestimation and underestimation in the actor-critic algorithms,a new learning method,randomly weighted triple Q-learning method is proposed.In addition,combining the new method with the soft actor critic algorithm,a new soft actor critic algorithm based on randomly weighted triple Q-learning is proposed.This algorithm not only limits the Q estimation value near the real Q value,but also increases the randomness of the Q estimation value through randomly weighted method,so as to solve the problems of overestimation and underestimation of action value in the learning process.Experiment results show that,compared to the SAC algorithm and other currently popular deep reinforcement learning algorithms such as DDPG,PPO and TD3,the SAC-RWTQ algorithm has better performance on several Mujoco tasks on the gym simulation platform.

Microblog Rumor Detection Method Based on Propagation Path Tree Kernel Learning

XU Jian-min, SUN Peng, WU Shu-fang

Computer Science. 2022, 49 (6): 342-349. doi:10.11896/jsjkx.210400096

Abstract

PDF(2920KB) ( 813 )

References | Related Articles | Metrics

The rapid development of online social platforms such as microblog promotes the widespread propagation of various rumors information,thereby posing potential threats to social order.Rumor detection on microblog can effectively curb the spread of rumors and is of great significance for purifying the network environment and maintaining social stability.In view of the fact that the traditional rumor detection model only considers the characteristics of users,contents and communication statistics,and ignores the structural problem that the characteristics of users′ influence and emotional feedback increase with the forwarding and comment relationship in the process of rumor communication,a path tree kernel rumor automatic detection model based on the microblog information propagation tree is proposed in this paper.It embeds users’ influence,emotional feedback,contents into the nodes ofpropagation tree.By calculating the path similarity from the root node to the leaf node in propagation tree,the similarity between the microblog information propagation tree structure is obtained.Furthermore,the model uses the support vector machine classifier based on the propagation path tree kernel todetect microblog rumors.Experimental results show that the accuracy of the proposed model reaches 93.5%,which is better than that of the rumor detection models without considering the structure of propagation path.

Anomaly Detection Framework of System Call Trace Based on Sequence and Frequency Patterns

WEI Hui, CHEN Ze-mao, ZHANG Li-qiang

Computer Science. 2022, 49 (6): 350-355. doi:10.11896/jsjkx.210500031

Abstract

PDF(1887KB) ( 487 )

References | Related Articles | Metrics

The existing system call-based anomaly intrusion detection methods can’t accurately describe the behavior of the process by a single trace pattern.In this paper,the process behavior is modeled based on the sequence and frequency patterns of system call trace,and a data-driven anomaly detection framework is designed.The framework could detect both sequential and quantitative anomalies of the system call trace simultaneously.With the help of combinational window mechanism,the framework could realize offline fine-grained learning and online anomaly real-time detection by meeting different requirements of offline trai-ning and online detection for extracting trace information.Performance comparison experiments of unknown anomalies detection are conducted on the ADFA-LD intrusion detection standard dataset.The results show that,compared with the four traditional machine learning methods and four deep learning methods,the comprehensive detection performance of the framework improves by about 10%.

Key Agreement Scheme Based on Ocean Acoustic Channel

LIANG Zhen-zhen, XU Ming

Computer Science. 2022, 49 (6): 356-362. doi:10.11896/jsjkx.210400097

Abstract

PDF(2594KB) ( 486 )

References | Related Articles | Metrics

Aiming at the problem that underwater acoustic channel is vulnerable to various threats and attacks due to the uncertainty of marine environment,a key agreement scheme based on ocean acoustic channel is proposed.Firstly,the uncertainty of marine environment is modeled,and the expressions of calculated noise,multipath and Doppler parameter expressions are constructed,and the concept of interference factor of underwater acoustic channel based on Rényi entropy is proposed.Secondly,a Hash function based on Twisted Edwards elliptic curve equation is constructed for conducting identity authentication and extracting the initial key.Then,the typical sequence of piecewise initial keys is used as initial seed to generate piecewise Toeplitz matrix,and the matrix multiplication of Toeplitz matrix and the initial key are used to generate the label by piecewise operation,and securely transfer the initial key.Finally,the initial key is hashed again for privacy amplification and a final secure key generated.The correctness,robustness and confidentiality of the scheme are proved by the information theory,and the upper bound of the probabi-lity of success of the active attack is obtained.Simulation results demonstrate that when the initial information amount is 50 000 bit,the upper bound of the success rate of adversary’s active attack is 4.3×10^－23,and the key generation rate is 631 bit/s.Compared with existing schemes,the proposed scheme has obvious advantages in key generation rate and bit error rate.