Computer Science

Adjoint Observation Schemes and Software Implementation Framework for Autonomous Robots

XUE Yuanzhou, YANG Shuo, MAO Xinjun

Computer Science. 2023, 50 (7): 1-9. doi:10.11896/jsjkx.221200020

Abstract

PDF(1775KB) ( 2047 )

References | Related Articles | Metrics

Autonomous robot is a kind of cyber-physical system which operates in an open environment and can make and execute its behavior autonomously.It generates and executes effective plans according to the task.The dynamic change of environment state often results in that the planned behavior strategy is no longer applicable to the current environment and the results of behavior execution are not in line with expectations,thus affecting the task achievement of autonomous robots.The above problems pose challenges to both the behavioral decision-making and the construction of autonomous robot software.On the one hand,autonomous robots need to strengthen the observation of environment states and their changes during the execution of its behavior strategy,and adjust the behavior strategy in a timely and flexible manner based on the observation results,which improves the complexity of the robot’s observation scheme and behavioral decision-making algorithm.On the other hand,the complex interactions of observation,decision-making and behavior execution enhances the complexity of software component abstraction and data interaction,and how to abstract the functions of software components such as sensing,decision-making and actuating of robots and provide suitable software architecture has become an important challenge for the software construction of autonomous robots.In view of the above challenges,the idea of adjoint behavior of autonomous robots is proposed,the adjoint interaction between observation and actuating behavior is clearly defined,and two adjoint observation schemes,conditional adjoint observation scheme and objective adjoint observation scheme,are proposed according to different stages of behavior execution,so as to improve the sensing of environmental changes and decision-making adjustment ability of autonomous robots.Secondly,AutoRobot,an autonomous robot software development framework based on multi-agent system,is developed,which abstracts the robot’s sensors,actuators and planners into a set of autonomous software agents,and the agents implement the above adjoint observation schemes through autonomous decision-making and collaboration.The AutoRobot framework designs and packages a set of reusable software components for different agent types,which can effectively support the reuse and efficient development of autonomous robot software.Finally,an experimental analysis is carried out in the simulation environment,and the efficiency and effectiveness of task planning and execution based on the adjoint observation schemes are verified by comparing it with ROSPlan and DESPOT,two autonomous robot task planning and execution methods.

Test Cases Generation Techniques for Root Cause Location of Fault

DU Hao, WANG Yunchao, YAN Chenyu, LI Xingwei

Computer Science. 2023, 50 (7): 10-17. doi:10.11896/jsjkx.220700128

Abstract

PDF(2014KB) ( 1917 )

References | Related Articles | Metrics

Vulnerability root cause localization is an important stage of software debugging,and spectrum-based fault root cause localization method is a hot issue in software automation debugging research,but the effectiveness of the positioning depends to a large extent on the quality of the test cases.Test inputs of different types of software are poorly generalized,and randomly gene-rated test inputs lead to large errors in analysis results due to overfitting or too many confounding items,resulting in limited application scenarios of such techniques at present.In this paper,we propose a phased exploration method Dgenerate based on crash paths to address the test case generation problem and implement the prototype tool Dloc.First,we use binary staking to insert staking path information in the basic block during the input stage of program execution,and then classify the original test inputs into common and guided types based on this information.Then,we use the dynamic energy scheduling algorithm to explore crash-related paths to generate high-quality test cases.Finally,the test cases are executed in the original program and execution information is traced to effectively locate the root cause of program fault through statistical analysis.In this paper,15 real CVE vulnerabilities in six different types of software are selected for experiments,and the results show that test cases generated by Dloc can improve location efficiency by 75% on average compared to previous techniques,and Dloc can output the code fragments related to the root causes of defect in the top five positions with an accuracy of 87%,which verifies the feasibility and practicality of the method system in this paper.

Test Case Generation Based on Web Application Front-end Behavior Model

LIU Ziwen, YU Lijuan, SU Yixing, ZHAO Yao, SHI Zhu

Computer Science. 2023, 50 (7): 18-26. doi:10.11896/jsjkx.220900143

Abstract

PDF(2319KB) ( 1841 )

References | Related Articles | Metrics

Test case generation based on Web application front-end model is an important process of Web application testing,but most existing models for Web applications only focus on Web pages and their events,ignoring event triggering conditions and subsequent actions.Therefore,in order to describe the dynamic behavior of modern Web applications more accurately,this paper defines a new Web application front-end behavior model(FBM).Because there may be internal variables in the triggering conditions of transition in the model,that is,there are dependencies between transitions,which will make the generated test cases cannot be executed according to the input sequence,thus affecting the test results.Therefore,an optimized grouping genetic algorithm is proposed to automatically generate the feasible transition path(FTP).Considering the characteristics of FTP generation problem,the algorithm makes a reasonable design of chromosome initialization and fitness function,and adds a repair operator to adjust the individual length to generate FTP which satisfies the migration coverage.This paper also introduces an adaptive genetic operator and simulated annealing receiving mechanism to reduce the number of iterations,thus improving the solution speed.Experimental results show that the algorithm can effectively guarantee the feasibility and coverage of transition path on the basis of higher solution efficiency.

Rule-based Technique for Detecting Risky Dynamic Typing Code

CHEN Zhifei, HAO Yang, CHEN Lin, XIAO Liang

Computer Science. 2023, 50 (7): 27-37. doi:10.11896/jsjkx.221100244

Abstract

PDF(1846KB) ( 1794 )

References | Related Articles | Metrics

Python has experienced an explosive growth in application in recent years,especially in scientific computing and machine learning areas.While Python’s dynamically-typed nature provides developers with powerful programming abstractions,that same dynamic type system allows for type-related defects to accumulate in code bases.To address these issues,this paper investigates six types of risky dynamic typing code which could incur type-related defects.This paper formally describes the rule of each type of risky dynamic typing code and then proposes a rule-based technique for detecting them.We conduct a study on 25 real-world Python open-source software projects (with the total size of more than 945kLOC).The results show that risky dynamic typing code is widespread in open-source software projects and a single Python method could gather multiple instances of inconsistent variable types,and the rule-based detection technique achieves high detection accuracy and good performance.The technique for detecting risky dynamic typing code and findings from this work will provide a strong reference and support for healthy evolvement of dynamic typing feature and quality assurance of software projects.

Policy Optimization Scheme of Refresh and Duplication Combination Based on LDPC Read Delay

ZHANG Yaofang, LI Peixuan, XIE Ping

Computer Science. 2023, 50 (7): 38-45. doi:10.11896/jsjkx.220900179

Abstract

PDF(2570KB) ( 264 )

References | Related Articles | Metrics

Aiming at the problem of reliability degradation caused by the increase of the density and capacity of flash memory,an optimization scheme of refresh and copy combination strategy based on LDPC read delay is proposed.In general,the original strategy is to add LDPC code module to flash memory and use hard and soft decoding to correct data errors.The traditional refresh strategy is based on the original strategy,when the LDPC soft decoding fails to correct the error,the refresh strategy is used to correct the error.The scheme is based on the characteristics of LDPC soft decoding 7 quantitative levels,and takes this as the judgment condition,using the method of analysis and comparison to determine that the condition of refresh is level 3,the condition of the copy is level 5,and the two methods are reasonably applied in the LDPC soft decoding mode.Compared with the previous two strategies,the average response time of flash memory is reduced,and the read performance of flash memory is improved to a certain extent.Simulation is performed on the extended platform of the simulator disksim+ssd,and experimental results show that,the average response time of this scheme is 10% shorter than that of the original strategy,and its flash memory lifetime is prolonged compared to traditional refresh strategies.

Disease Diagnosis Prediction Algorithm Based on Contrastive Learning

WANG Mingxia, XIONG Yun

Computer Science. 2023, 50 (7): 46-52. doi:10.11896/jsjkx.230200216

Abstract

PDF(2321KB) ( 412 )

References | Related Articles | Metrics

Disease diagnosis prediction aims to use electronic health data to model disease progression patterns and predict the future health status of patients,and is widely used in assisting clinical decision-making,healthcare services and other fields.In order to further explore the valuable information in the medical records,a disease diagnosis prediction algorithm based on contrastive learning is proposed.Contrastive learning provides self-supervised training signals for the model by measuring the similarity between samples,which can improve the information capture ability of the model.The proposed algorithm excavates the common knowledge between similar patients through contrastive training,and enhances the ability of the model to learn patient representations.In order to capture more comprehensive common information,the information of similar groups of the target patient is further explored as auxiliary information to characterize the health status of the target patient.Experimental results on the public dataset show that compared with the Retain,Dipole,LSAN and GRASP algorithms,the proposed algorithm improves AUROC and AUPRC of the readmission prediction task by more than 2.9% and 8.1% respectively,and Recall@10 and MAP@10 of the diagnosis prediction task by 2.1% and 1.8%,respectively.

Dually Encoded Semi-supervised Anomaly Detection

LI Hui, LI Wengen, GUAN Jihong

Computer Science. 2023, 50 (7): 53-59. doi:10.11896/jsjkx.220900027

Abstract

PDF(2190KB) ( 335 )

References | Related Articles | Metrics

Anomaly detection is a hot topic that has been widely studied in the field of machine learning and plays an important role in industrial production,food safety,disease monitoring,etc.The latest anomaly detection methods mostly jointly train semi-supervised detection models based on a small number of available labeled samples and many unlabeled samples.However,these existing semi-supervised anomaly detection models mostly use deep learning frameworks.Due to the lack of enough feature information on low-dimensional data sets,it is difficult to learn accurate data boundaries,resulting in insufficient detection perfor-mance.To solve this problem,a dually encoded semi-supervised anomaly detection(DE-SAD)model is proposed.DE-SAD can make full use of a small amount of available labeled data and a large amount of unlabeled data for semi-supervised learning,and learn more accurate implicit manifold distribution of normal data through the dually encoded stage constraint,thus effectively magnifying the gap between normal data and abnormal data.DE-SAD shows excellent ano-maly detection performance on multiple anomaly detection datasets from different fields,especially on low-dimensional data,and its AUROC is up to 4.6% higher than the current state-of-the-art methods.

Event Recommendation Method with Multi-factor Feature Fusion in EBSN

SHAN Xiaohuan, SONG Rui, LI Haihai, SONG Baoyan

Computer Science. 2023, 50 (7): 60-65. doi:10.11896/jsjkx.220900036

Abstract

PDF(2508KB) ( 262 )

References | Related Articles | Metrics

Event-based social network(EBSN) is a new kind of complex heterogeneous social network,the personalized event re-commendation in it has certain application value.In recent years,with the rapid development of EBSN,the problem of information overload for event recommendation has been solved by data mining technology.However,it will reduce the accuracy of event re-commendation by only using a single feature attribute or a small number of linear combinations for calculation,and predefining fixed weights.In addition,most approaches ignore the influence of user feedback information on subsequent recommendation.Aiming at the above problems,an event recommendation method fusing multi-factor features is proposed,which consists of two phases.In the query preprocessing phase,the events,historical users and their relationships in EBSN are abstracted as a directed he-terogeneous graph,and the feature information of nodes and edges is extracted for auxiliary storage.A relatively small candidate set is obtained by filtering invalid nodes and edges with the auxiliary data.According to the query context,the query semantics are transformed into the query graphs.In the online query phase,it combines the characteristics of potential friends,event-based collaborative filtering and users’ interests to recommend,and also receives feedback from users on whether they accept the event as a reference factor for subsequent recommendations.Large number of experiments on real datasets and simulated datasets verify the accuracy and user satisfaction of the proposed method in EBSN event recommendation.

Variational Continuous Bayesian Meta-learning Based Algorithm for Recommendation

ZHU Wentao, LIU Wei, LIANG Shangsong, ZHU Huaijie, YIN Jian

Computer Science. 2023, 50 (7): 66-71. doi:10.11896/jsjkx.220900125

Abstract

PDF(2130KB) ( 317 )

References | Related Articles | Metrics

Meta-learning methods have been introduced into recommendation algorithms in recent years to alleviate the problem of cold start.The existing meta-learning algorithms can only improve the ability of the algorithm to deal with a set of statically distributed data sets(tasks).When faced with multiple data sets subject to non-stationary distribution,the existing models often have negative knowledge transfer and catastrophic forgetting problems,resulting in a significant decline in algorithm recommendation performance.This paper explores a recommendation algorithm based on variational continuous Bayesian Meta-learning(VC-BML).Firstly,the algorithm assumes that the meta parameters follow the dynamic mixed Gaussian model,which makes it have a larger parameter space,improves the ability of the model to adapt to different tasks,and alleviates the problem of negative knowledge transfer.Then,the number of task clusters in VC-BML is flexibly determined by Chinese restaurant process(CRP),which enables the model to store knowledge of different task distributions in different mixed components and invoke this know-ledge when similar tasks occur,helping to alleviate the catastrophic forgetting problem in traditional algorithms.To estimate the posterior probabilities of model parameters,a more robust structured variational reasoning method is used to approximate the posterior values to avoid forgetting knowledge.Finally,VC-BML outperforms the baselines on all four datasets with non-stationary distributions.Compared with point-based estimation methods,VC-BML improves the robustness of the model and helps alle-viate the catastrophic forgetting problem.

Unsupervised Feature Selection Algorithm Based on Dual Manifold Re-ranking

LIANG Yunhui, GAN Jianwen, CHEN Yan, ZHOU Peng, DU Liang

Computer Science. 2023, 50 (7): 72-81. doi:10.11896/jsjkx.221000143

Abstract

PDF(4183KB) ( 279 )

References | Related Articles | Metrics

High dimensional data is often encountered in many data analysis tasks.Feature selection techniques aim to find the most representative features from the original high-dimensional data.Due to the lack of class label information,it is much more difficult to select suitable features in unsupervised learning scenarios than in supervised scenarios.Traditional unsupervised feature selection methods usually score the features of samples according to certain criteria in which samples are treated indiscriminately.However,these approaches cannot capture the internal structure of data completely.The importance of different samples should vary.There is a dual relationship between weight of sample and feature that will influence each other.Therefore,an unsupervised feature selection algorithm based on dual manifold re-ranking(DMRR) is proposed in this paper.Different similarity matrices are constructed to depict the manifold structures on samples and samples,features and features,and samples and features respectively.Then manifold re-ranking is carried out by combining the initial scores of samples and features.By comparing DMRR with three original unsupervised feature selection algorithms and two unsupervised feature selection post-processing algorithms,experimental results verify that importance information of different samples and the dual relationship between sample and feature are helpful to achieve better feature selection.

Method for Correlation Data Imputation Based on Compressed Sensing

REN Bing, GUO Yan, LI Ning, LIU Cuntao

Computer Science. 2023, 50 (7): 82-88. doi:10.11896/jsjkx.220600209

Abstract

PDF(2237KB) ( 220 )

References | Related Articles | Metrics

The phenomenon of missing data occurs frequently during the acquisition and transfer of data,and improper handling of missing data sets can adversely affect subsequent data mining efforts.In order to fill the missing data set more effectively,a method for data imputation based on compressed sensing is proposed for correlation data.First,the problem of missing data imputation is transformed into a sparse vector recovery problem under the compressed sensing framework.Second,a specialized sparse representation base is constructed for correlation data,so the data sparsity can be better realized.Finally,the fast iterative weighted thresholding algorithm(FIWTA) is proposed,which is refined based on the fast iterative shrinkage-thresholding algorithm (FISTA).The proposed algorithm adopts a new iterative weighted method and introduces a restart strategy,which greatly improves the convergence of the algorithm and the reconstruction accuracy of the data.Simulation results show that the proposed algorithm is able to fill the missing data efficiently,and both the reconstruction success rate and the reconstruction speed are improved compared with the traditional fast iterative shrinkage-thresholding algorithm.Meanwhile,even when the sparse transformation of the data is less effective,imputation of missing data sets can still be accomplished with better robustness.

Block Sparse Symmetric Nonnegative Matrix Factorization Based on Constrained Graph Regularization

LIU Wei, DENG Xiuqin, LIU Dongdong, LIU Yulan

Computer Science. 2023, 50 (7): 89-97. doi:10.11896/jsjkx.220500050

Abstract

PDF(3278KB) ( 243 )

References | Related Articles | Metrics

The existing algorithms based on symmetric nonnegative matrix factorization(SymNMF) are mostly rely on initial data to construct affinity matrices,and neglect the limited pairwise constraints,so these methods are unable to effectively distinguish similar samples of different categories or learn the geometric features of samples.To solve the above problems,this paper proposes a block sparse symmetric nonnegative matrix factorization based on constrained graph regularization(CGBS-SymNMF).Firstly,the constrained graph matrix is constructed by prior information,which is used to guide the clustering indicator matrix to distinguish different clusters of samples with high similarity.Secondly,pairwise constraint propagation by semidefinite programming(PCP-SDP) is introduced to learn a new sample graph mapping matrix by using pairwise constraints.Finally,a dissimilarity matrix is constructed by cannot-link constraints,which is used to guide a block sparse regular term for enhancing the anti-noise capability of the model.Experimental results demonstrate a higher clustering accuracy and stability of the proposed algorithm.

Exploring Station Spatio-Temporal Mobility Pattern:A Short and Long-term Traffic Prediction Framework

SHEN Zhehui, WANG Kailai, KONG Xiangjie

Computer Science. 2023, 50 (7): 98-106. doi:10.11896/jsjkx.220900109

Abstract

PDF(2971KB) ( 250 )

References | Related Articles | Metrics

With the technological development of intelligent transportation system and the surging spatio-temporal data in urban,the demand for public services is increasingly emphasized.As a vital part of urban transportation,public transportation also faces enormous challenges,and the spatio-temporal prediction task in transportation network is the core of the solutions for various traffic problems.Mobility pattern in traffic can reflect the travel behaviors of people and their rules.In most studies on traffic prediction task,the importance of mobility pattern is neglected.In view of the problem of existing work,a multi-pattern traffic prediction framework,MPGNNFormer,is proposed,in which based-graph neural network deep clustering method is used to extract mobility patterns of stations,and a Transformer-based spatio-temporal prediction model is designed to learn temporal dependence and spatial dependence of stations and to improve the computational efficiency.Then,a series of experiments are conducted on real bus dataset for evaluation and testing,including analysis of mobility patterns and comparison of prediction results.Finally,experimental results prove the efficacy of proposed method in the short and long-term traffic prediction of traffic networks,and its sca-lability is discussed.

Review of 3D Object Detection for Autonomous Driving

HUO Weile, JING Tao, REN Shuang

Computer Science. 2023, 50 (7): 107-118. doi:10.11896/jsjkx.220700090

Abstract

PDF(2785KB) ( 370 )

References | Related Articles | Metrics

In recent years,with the rapid development of autonomous driving,3D object detection technology as the core of perception systems has received more and more attention and become a hot research direction.At the same time,the wide application of deep learning has made a great breakthrough in 3D object detection technology recently.A large number of excellent algorithms have emerged.This paper systematically summarizes 3D object detection methods for the autonomous driving field and divides the existing algorithms into three categories according to sensor types:image-based 3D object detection,LiDAR-based 3D object detection,and multi-sensor-based 3D object detection.After that,it analyzes the advantages and disadvantages of the three methods in detail.The LiDAR-based 3D object detection algorithms are thoroughly investigated and subdivided.Then it introduces the commonly used 3D object detection datasets in autonomous driving,including KITTI,nuScenes,and Waymo Open Dataset,and compares the performance of the latest 3D object detection algorithms on different datasets.Finally,the future research direction of 3D object detection technology is discussed.

Constructing Combined Quadratic h-Bezier Curves with Monotone Curvature

LI Lin, XIE Bin, HAN Liwen

Computer Science. 2023, 50 (7): 119-128. doi:10.11896/jsjkx.220800024

Abstract

PDF(2949KB) ( 238 )

References | Related Articles | Metrics

The h-Bézier curves(h>0),also known as Pólya curves,share many excellent properties with classical Bézier curves(h=0).In this paper,the necessary and sufficient conditions for quadratic h-Bézier curves with monotonic curvature and its construction algorithm are studied.First,the sufficient and necessary conditions for quadratic h-Bézier curves with monotone curvature are obtained by discussing the existence of the extremes of the curves.By introducing curvature critical circles of the curvature,the monotony for the curvature of the quadratic h-Bézier curve can be directly verified by checking whether the middle control point of the curve is on or inside the curvature critical circle.Two algorithms for construct quadratic h-Bezier curves with monotonic curvature are obtained,ensuring that the curves can have monotonically decreasing or increasing curvature by adjusting the shape parameter h.Secondly,the G² smooth blending of two quadratic h-Bézier curves is studied.Based on the analysis of the properties of the quadratic h-Bézier curves,the shoulder point of the second curve is selected to join with the end point of the first curve,and the necessary and sufficient conditions are obtained and the influence of parameters on the shape of the blending curve is discussed.Finally,the combined quadratic h-Bézier curve with decreasing(or increasing) is constructed.The numerical examples show the modeling advantage and flexibility of the combined quadratic h-Bézier curve.

Arbitrary Image Style Transfer with Consistent Semantic Style

YAN Mingqiang, YU Pengfei, LI Haiyan, LI Hongsong

Computer Science. 2023, 50 (7): 129-136. doi:10.11896/jsjkx.220700008

Abstract

PDF(5451KB) ( 307 )

References | Related Articles | Metrics

The goal of image style transfer is to synthesize an output image by transferring the style of the target image to a given content image.There are a large number of image style transfer works,but the stylization results ignore the manifold distribution of different semantic regions of the content image.At the same time,most methods use global statistics(for example,Gram matrix or covariance matrix) to achieve the matching of style feature to content feature.There are inevitable issues of content loss,style leakage,and the presence of artifacts,resulting in inconsistent stylized results.Aiming at the above problems,a self-attention mechanism-based progressive manifold feature mapping module(MFMM-AM) is proposed to coordinately match features between related content and style manifolds.Exact histogram matching(EHM) is applied to achieve higher-order distribution ma-tching of style and content feature maps,reducing the loss of image information.Finally,two contrastive losses are introduced to learn human beings using the external information of large-scale style datasets perceived style information that makes the color distribution and texture patterns of stylized images more reasonable.Experimental results show that,compared with the existing typical arbitrary image style transfer methods,the proposed network greatly bridges the gap between human-created artworks and AI-created artworks,and can generate visually more harmonious and satisfying artistic images.

Study on Single Background Object Detection Oriented Improved-RetinaNet Model and Its Application

ZHOU Bo, JIANG Peifeng, DUAN Chang, LUO Yuetong

Computer Science. 2023, 50 (7): 137-142. doi:10.11896/jsjkx.220500066

Abstract

PDF(2455KB) ( 235 )

References | Related Articles | Metrics

Object detection algorithms based on deep learning have been fully promoted and applied in the field of industrial defect detection,but few algorithms are suitable for single background in industrial detection scenarios.This paper takes the industrial detection scene with a large number of simple repeated backgrounds as the starting point,and makes the following improvements to the RetinaNet algorithm:1)introduce the difficult negative sample mining strategy to reduce the impact of a large number of simple repeated negative samples on the model fitting positive samples;2)an adaptive ignoring sample selection strategy is designed to avoid mixing samples with high intersection ratios of positive samples into negative samples to confuse model training;3)the classification sub-network of RetinaNet is simplified,and the risk of overfitting after model improvement is reduced.Compared with the RetinaNet algorithm,the improved method improves the recall rate by 14.1% and 1.8%,and the precision rate by 3.6% and 0.4% respectively on the public PCB missing hole dataset and the self-built LED bubble dataset,indicating that the improved method can significantly improve the level of object detection in a single background.

Medical Ultrasound Image Super-resolution Reconstruction Based on Video Multi-frame Fusion

ZHAO Ran, YUAN Jiabin, FAN Lili

Computer Science. 2023, 50 (7): 143-151. doi:10.11896/jsjkx.220700232

Abstract

PDF(2320KB) ( 322 )

References | Related Articles | Metrics

Medical ultrasound imaging is one of the most widely used methods in clinical diagnosis.However,the resolution and contrast of ultrasound images are low,and the noise is serious.Image Super-resolution is widely used to improve the quality of ultrasound images.However,the exsisting studies lack full use of complementary information in ultrasound video frames,so the results are not ideal.To solve this problem,this paper proposes a super-resolution method of medical ultrasound images based on video multi-frame fusion.Firstly,it builds a multi-frame fusion model based on convolutional neural network for ultrasound images.The model obtains the fused image with rich information by feature fusion of continuous multi-frame images.Then,it builds a lightweight image super-resolution model based on data-free knowledge distillation.The model trains fused feature images to obtain a teacher super-resolution model.To obtain the final high-quality medical ultrasound image,it creates a lightweight student super-resolution model using the trained teacher model and training data from the generative adversarial network.Finally,the proposed method is tested on large ultrasound datasets,using two objective image evaluation indicators and results on image classification to evaluate its performance.The results show that compared with other methods,the proposed method not only improves the resolution of ultrasound images,but also obtains ultrasound images with more information and higher contrast.The recognition accuracy of the super-resolution image obtained by this method in the classification network can reach 97.30%,which is much higher than that of previous approaches and can increase productivity and the precision of clinical diagnoses.

Incremental Pose Graph Segmentation Algorithm Based on Camera Orientation Change

FAN Hanqi, WANG Shaojing

Computer Science. 2023, 50 (7): 152-159. doi:10.11896/jsjkx.220400166

Abstract

PDF(3702KB) ( 252 )

References | Related Articles | Metrics

In camera trajectory estimation,pose graph is one of the most widely used tools to reduce the cumulative error.How-ever,the scale of the pose graph grows as the cameras move,which would render trajectory estimation impossible in the real-time required applications like AR/VR(augmented reality/virtual reality).To reduce the size of the pose graph optimization,this paper proposes an algorithm that segments the trajectory of cameras incrementally by the change of orientation.The orientation changes significantly where the trajectory is segmented by the proposed algorithm.Efficiency is improved by optimizing the cameras with obvious orientation changes instead of the entire trajectory.In addition,the starting camera and the ending camera of the trajectory segment are utilized as references to estimate different poses for each camera within the trajectory segment,followed by the weighted average method is used on its different poses to solve the final pose.Which not only avoids a large amount of computation of nonlinear optimization,but also reduces the influence of noise to achieve high accuracy.Experiments on EuRoC,TUM,and KITTI datasets show that the proposed scheme reduces the size of the pose graph optimization and ensures the accuracy of trajectory.

Unsupervised Domain Adaptive Pedestrian Re-identification Based on Counterfactual AttentionLearning

DAI Xuesong, LI Xiaohong, ZHANG Jingjing, QI Meibin, LIU Yimin

Computer Science. 2023, 50 (7): 160-166. doi:10.11896/jsjkx.220600153

Abstract

PDF(2051KB) ( 344 )

References | Related Articles | Metrics

Most of the existing unsupervised domain adaptive pedestrian re-identification methods combine clustering-based pseudo-label prediction with feature fine-tuning.Due to the differences between domains,incorrect pseudo-labels are generated during the clustering process,making pseudo-labels unreliable to a certain extent,misleading feature representation learning,and affec-ting the performance of domain-adaptive models.First,a novel unsupervised domain adaptive network based on counterfactual attention learning is designed,which guides and optimizes the training process by measuring the quality of attention learning,prompting the model to focus on more accurate attention features and reducing the generation of noisy pseudo-labels.Secondly,a noisy samples optimization method based on uncertainty evaluation is proposed.By measuring the inconsistency level between the output features of the student model and the teacher model,as the uncertainty distribution of pedestrian samples in the target domain.The teacher model and the student model are both constructed based on the average teacher method.The uncertainty of the sample is used to reasonably weight each part of the overall loss of the network,and the erroneous influence of the sample with high uncertainty on the overall loss of the model is corrected,and the recognition performance of the target domain is further improved.Experimental data show that the proposed method significantly improves the experimental results in both the source domain DukeMTMC-reID/Market-1501 and the target domain Market-1501/DukeMTMC-reID,with mAP and Rank-1 reaching 82.9%,93.6% and 71.8%,84.4%,respectively.

Automated Reasoning Techniques for Solving Combinatorial Mathematical Problems:A Survey

HUANG Pei, LIU Minghao, MA Feifei, ZHANG Jian

Computer Science. 2023, 50 (7): 167-175. doi:10.11896/jsjkx.221000251

Abstract

PDF(1908KB) ( 298 )

References | Related Articles | Metrics

Automated reasoning is a symbolic algorithmic technique that aims to simulate the logical reasoning ability of human.The overall goal is to mechanize different forms of reasoning with a computer system.Although the theoretical framework of the field has not yet supported the simulation of the full range of human reasoning capabilities,developments in the field have reached a point where automated reasoning programs are being used by researchers to attack open problems in mathematics and logic and provide important applications in computing science.This paper briefly reviews common approaches of automatic reasoning in dealing with open problems in combinatorial mathematics and highlights the latest developments in this field in China and abroad.Then the strengths and weaknesses of various approaches are analyzed,and reasoning strategies that have emerged in recent years to enhance the trustworthiness of automated reasoning results are introduced.Finally,future research directions and challenges are discussed.

Survey of Analysis and Solutions for Multi-UAV Cooperative Mission Planning Problem Under Multi-constraint Conditions

HU Jiawei, JIA Zequn, SUN Yantao, LIU Qiang

Computer Science. 2023, 50 (7): 176-193. doi:10.11896/jsjkx.220700066

Abstract

PDF(2983KB) ( 509 )

References | Related Articles | Metrics

Multi-UAV cooperative mission planning is the key technology for the intelligent development of UAV swarms at this stage,and task allocation and trajectory planning are the core parts of UAV mission planning technology.As there are numerous and coupled planning elements,it is important to find a solution strategy that reduces both the coupling and complexity of the problem.Firstly,starting from problem modeling,a general model of task planning problem is established,its common constraints and evaluation indicators are summarized,and its problem solving framework is emphatically analyzed.Secondly,common models and solving algorithms of task assignment problem are expounded from the centralized and distributed perspectives,respectively.Afterwards,single-UAV path planning and path smoothing algorithms are discussed,and cooperative methods with multi-UAV space-time cooperative constraints are summarized.In addition,complex coupling factors in the problem solving process are summarized from the perspectives of constraint coupling,sub-problem coupling and hierarchical structure coupling,and the corres-ponding decoupling strategies are discussed emphatically.Finally,the development trend of multi-UAV cooperative mission planning problem in the future is analyzed.

Improved NSGA-III Based on Kriging Model for Expensive Many-objective Optimization Problems

GENG Huantong, SONG Feifei, ZHOU Zhengli, XU Xiaohan

Computer Science. 2023, 50 (7): 194-206. doi:10.11896/jsjkx.220600186

Abstract

PDF(3460KB) ( 299 )

References | Related Articles | Metrics

In many real world multi-objective optimization problems(MOP),the cost of physical experiments or numerical simulations for fitness evaluation is very expensive,which poses a great challenge to most existing multi-objective evolutionary algorithmEAs).Therefore,this paper proposes an improved reference point guided evolution optimization algorithm assisted by Kriging model to solve expensive many-objective optimization.Specifically,according to the distribution characteristics of the target spatial population,the reference points are selected to guide the evolution of the population to reach the balance of exploration and exploitation.The proposed surrogate-assisted evolution algorithm(SAEA) uses Kriging method to approximate each objective function without the need for the original expensive function evaluation to reduce computational cost.In model management,an infill sampling criterion is adopted to improve population convergence and algorithm optimization efficiency by evaluating convergence and diversity indexes to determine the appropriate sampling strategy for re-evaluation with expensive objective functions.The effectiveness and superiority of the proposed algorithm are proved by the empirical research on the benchmark problems with more than three objectives.

Self-supervised Dynamic Graph Representation Learning Approach Based on Contrastive Prediction

JIANG Linpu, CHEN Kejia

Computer Science. 2023, 50 (7): 207-212. doi:10.11896/jsjkx.220500093

Abstract

PDF(1903KB) ( 297 )

References | Related Articles | Metrics

In recent years,graph self-supervised learning represented by graph contrastive learning has become a hot research to-pic in the field of graph learning.This learning paradigm does not depend on node labels and has good generalization ability.However,most of the existing graph self-supervised learning methods use static graph structures to design learning tasks,such as learning node-level or graph-level representations based on structural contrast,without considering the dynamic information of graph over time.To address this problem,the paper proposes a self-supervised dynamic graph representation learning method based on contrastive prediction(DGCP),which utilizes a contrastive loss inducing the embedding space to capture the most useful information for predicting future graph structures.Firstly,each temporal snapshot graph is encoded using a graph neural network to obtain its corresponding node representation matrix.Then,an autoregressive model is used to predict node representations in the next temporal snapshot graph.Finally,the model is trained end-to-end by using the contrastive loss and sliding window me-chanism.Experimental results on real graph datasets show that DGCP outperforms baseline methods on the link prediction task.

Short-term Subway Passenger Flow Forecasting Based on Graphical Embedding of Temporal Knowledge

MAO Huihui, ZHAO Xiaole, DU Shengdong, TENG Fei, LI Tianrui

Computer Science. 2023, 50 (7): 213-220. doi:10.11896/jsjkx.220600120

Abstract

PDF(2238KB) ( 396 )

References | Related Articles | Metrics

Subway short-term passenger flow forecasting is an essential component in urban subway operation,and it aims to forecast the passenger flow of subway stations in a short time in the future.Aiming at the problem that the existing methods fail to make full use of the passenger flow information of stations,a short-term subway passenger flow forecasting method based on temporal knowledge graph embedding combined with residual network and long short-term memory network is proposed,which is called TKG-ResLSTM.First,we use subway passenger flow data to construct a temporal knowledge graph of subway passenger flow,and apply the graphical embedding of temporal knowledge to obtain the dynamic patterns of subway stations passenger flow.Then,the extracted dynamic patterns of passenger flow are converted into dynamic similarity matrices and applied to theforecasting architecture of subway passenger flow based on deep learning to complete the subway passenger flow forecasting task.Finally,experimental evaluations are carried out at time granularities of 10 min,15 min,and 30 min using the Beijing subway and city A subway passenger flow datasets,respectively.Experimental results show that TKG-ResLSTM can effectively extract the dynamic patterns of subway stations passenger flow.Without using external information,TKG-ResLSTM reduces the root mean square error of forecasting by 0.41 compared with ResLSTM in the time granularity of 10 min of the Beijing subway dataset.

New Word Detection Based on Branch Entropy-Segmentation Probability Model

ZHU Yuying, GUO Yan, WAN Yizhao, TIAN Kai

Computer Science. 2023, 50 (7): 221-228. doi:10.11896/jsjkx.220700074

Abstract

PDF(2211KB) ( 264 )

References | Related Articles | Metrics

As a basic task of Chinese natural language processing,new word detection is crucial for improving the performance of various downstream tasks.This paper proposes a new word detection method based on branch entropy and segmentation probabi-lity.The method firstly generates a candidate word set from the text based on branch entropy,and then calculates the segmentation probability of each candidate,so as to filter out the noisy word.Two different models are proposed to respectively deal with situations whether or not there are annotated corpus related to the text to be processed.In the absence of related segmented corpus,the multi-criteria Transformer-CRF model is trained using general segmented benchmark data sets.A key-value based memory neural network is introduced to fully extract the wordhood information if there is field-specific segmented corpus.Experimental results show that the multi-criteria Transformer-CRF model has a MAP of 54.00% of legal texts in the top 900 resulted words,which is 2.15% higher than that of the unsupervised method.As with segmented legal corpus,the performance of the key-value memory neural network further exceeds the former model,has an improvement of 3.43%.

Recognition Method of Component Names in Patent Documents Based on the Algorithm of Word Frequency Difference and Library of Left-segmentation Words

KONG Jiabin, LYU Jianwen, LIU Jiangnan, DU Wenxuan

Computer Science. 2023, 50 (7): 229-236. doi:10.11896/jsjkx.220500068

Abstract

PDF(2580KB) ( 190 )

References | Related Articles | Metrics

Mechanical patent literature contains a large amount of domain knowledge where component names exist as information units.Being flexible and changeable,the word formatting of component name represents the characteristics of uniqueness,complexity and lesser-known expressions.The challenge of accurate recognition of component names by computers becomes an obstacle to patent knowledge mining.In order to propose an efficient method to recognize component names,the features of word formation in patent text statements are analyzed and extracted.Starting with external words related to component names,characters on the left side of the appended drawing reference signs(ADRS) are identified.Accordingly,candidate names are automatically retrieved from texts,and the set of candidate names are constructed.An algorithm of word frequency difference is proposed to filter redundant characters in the set of candidate names.By building left-segmentation library(LSL) dynamically,redundant characters which are not filtered are further eliminated.Based on cross-over experiment,the influence of character frequency difference prior threshold(CFDV-Ⅰ),word frequency threshold(LSWF) and character frequency difference threshold(CFDV-Ⅱ) on recognition result is tested and analyzed.Furthermore,a three-stage comprehensive method for recognizing component names from patent documents in mechanical field is proposed.Finally,the method has been proved to be effective and efficient by comparing the results of experiments.

Anomaly Detection Method Based on Context Information Fusion and Noise Adaptation

HENG Hongjun, ZHOU Wenhua

Computer Science. 2023, 50 (7): 237-245. doi:10.11896/jsjkx.220700078

Abstract

PDF(2235KB) ( 246 )

References | Related Articles | Metrics

The data collected by field devices such as sensors and actuators in cyber-physical systems(CPSs) contains complex context information.To extract and fuse context information in data and reduce the interference caused by noise,an anomaly detection method based on context information fusion and noise adaptation is proposed.In this method,an encoder-decoder network composed of adaptive denoising encoder,context information encoder and decoder is designed to model the state-space model of CPSs.The adaptive denoising encoder generates adaptive noise by fitting the distribution of noise in the data during the training process,and adds the adaptive noise to the sensor data of the training data,so as to improve the robustness of the encoder-decoder network,reduce the interference caused by noise,and force the adaptive denoising decoder to learn the system hidden state with stronger generalization.Context information encoder uses long-short term memory(LSTM) and convolutional neural networks(CNN) to extract temporal context information and spatial context information in the data window,and uses self-attention mechanism to fuse these two types of context information with system hidden state,so as to obtain the fusion result,which is used to infer the system hidden state at the current moment,so as to increase the amount of information of system hidden state at the current moment.The decoder can decode the corresponding sensor data more accurately by using the above system hidden states.After the encoder-decoder network training is completed,the system hidden state and the decoded sensor data are obtained,and the anomaly score is calculated based on the unscented Kalman filter algorithm.Experimental results on two actual CPSs datasets of SWaT and PUMP show that the F1 value of the proposed method is better than other comparison methods,which verifies its effectiveness.

Task Scheduling Strategy for Energy Consumption Optimization of Cloud Data Center Based on Improved Particle Swarm Algorithm

LIU Chenwei, SUN Jian, LEI Bingbing, XU Tao, WU Zhuiwei

Computer Science. 2023, 50 (7): 246-253. doi:10.11896/jsjkx.220900176

Abstract

PDF(2589KB) ( 226 )

References | Related Articles | Metrics

With the development of cloud computing,energy consumptionhas increased dramatically,which further limits the improvement of the overall performance of the cloud data center,and thus the energy consumption issue has attracted the attention of industry and academia.Meanwhile,traditional particle swarm optimization algorithm(PSO) is widely used to solve data center task scheduling problems,but it has the shortcomings of slow convergence and low accuracy,and it is easy to ignore the cluster energy consumption problem.A chaotic mapping adaptive particle swarm optimization algorithm based on opposition-based lear-ning(OAPSO) is proposed.Firstly,the initial population is generated by the method of opposition-based learning,which makes the particles more evenly distributed in the initial solution space and improves the quality of the initial population.Secondly,a nonlinear decreasing dynamic inertia weight stra-tegy is introduced into the particle updating mode to change the particle optimization ability,so as to balance the local search and global search and avoid the algorithm falling into the local optimal.Thirdly,the chaotic mapping strategy is introduced to generate new solutions by perturbation and mutation at the optimal location,which improves the ability of the algorithm to jump out of the local optimal.Finally,the proposed algorithm is verified by experiments on the Cloudsim platform,and the results show that,compared to PSO,OBL_ TP_PSO and SAPSO,OAPSO algorithm has higher resource utilization and better energy-saving effect.

Cloud Platform Load Prediction Method Based on Temporal Convolutional Network

LI Yinghao, GUO Haogong, LIU Panpan, XIANG Yihao, LIU Chengming

Computer Science. 2023, 50 (7): 254-260. doi:10.11896/jsjkx.220500036

Abstract

PDF(3431KB) ( 268 )

References | Related Articles | Metrics

Aiming at the problems of highly non-stationary cloud platform resource load data and low prediction accuracy due to random noise,a cloud platform resource load prediction method is proposed by combining signal decomposition and deep learning technologies.Firstly,the original data is decomposed using empirical mode decomposition(EMD) method to obtain multiple IMF components;then a prediction model based on temporal convolutional network(TCN) is constructed to predict the IMF components separately;finally,the prediction results are combined to obtain the final prediction value.The proposed method is compared with traditional prediction methods and deep learning prediction methods,and a comparative experiment is carried out on Alibaba’sopen source data center resource monitoring log data set.Experimental data results show that the prediction errors of the proposed method reduces by 36.75%,23.5%,24.44%,and 24.53% compared with ARIMA,Bi-LSTM,GRU and TCN,respectively,and the prediction results have the best accuracy.

Deep Learning-based Algorithm for Active IPv6 Address Prediction

LI Yuqiang, LI Linfeng, ZHU Hao, HOU Mengshu

Computer Science. 2023, 50 (7): 261-269. doi:10.11896/jsjkx.220700076

Abstract

PDF(3219KB) ( 288 )

References | Related Articles | Metrics

The huge address space of IPv6 makes it difficult to achieve a global IPv6 address scan based on the existing network speed and hardware computing power.Fast IPv6 address scanning can be achieved by using address generation algorithms to predict the possible IPv6 addresses in the network and subsequently using the predicted addresses as the targets of scanning.This paper explores potential allocation patterns by analyzing IPv6 address structures and allocation methods,and proposes a deep learning-based algorithm 6LMNS to predict potentially active IPV6 addresses by combining existing traditional language models and target generation algorithms.6LMNS first constructs IPv6 address word vector spaces with certain semantic relationships through the address vector space mapping model Add2vec.Subsequently,the language training model GPT-IPv6 is constructed based on Transformers to estimate the probability distribution of IPv6 address word sequences.Finally,nucleus sampling is introduced instead of traditional greedy search decoding to complete the generation of active addresses.It is verified that the addresses generated by 6LMNS have better diversity as well as higher activity rate compared with other language models and target generation algorithms.

APPOINTER:Adaptive Network Telemetry Path Orchestration Method Based on Cooperative Migration Evolution

HAO Bingwei, CUI Yunhe, QIAN Qing, SHEN Guowei, GUO Chun

Computer Science. 2023, 50 (7): 270-277. doi:10.11896/jsjkx.220500274

Abstract

PDF(3008KB) ( 223 )

References | Related Articles | Metrics

The increasingly large,complex and high-speed network makes the traditional network measurement technology cannot meet the current demand of intelligent network control.As a new network measurement technology,network telemetry can provide fine-grained and accurate session-level or message-level telemetry information.The existing network telemetry solutions do not consider the network state when deploying the network telemetry path,and deploy the network telemetry path in a static manner.These approaches cannot adapt to the dynamic and unreliable nature of the network.If the routing path that transfers the network telemetry packets is facing with network attacks or saturated,the network telemetry packets will be lost,and the reliabi-lity of network telemetry will decreased.In addition,the existing network telemetry methods are usually implemented by traversing all links,causing large telemetry redundancy and relatively low probe packet payload.To solve the above problems,this paper proposes APPOINTER,an adaptive network telemetry path scheduling method based on cooperative migration evolution.APPOINTER calculates the optimal network telemetry path that can traverses all network devices to forward telemetry messages based on network state information.Experimental results show that APPOINTER enhances the reliability of network telemetry,effectively avoids telemetry redundancy,and improves telemetry efficiency.

Stackelberg Model Based Distributed Pricing and Computation Offloading in Mobile Edge Computing

CHEN Xuzhan, LIN Bing, CHEN Xing

Computer Science. 2023, 50 (7): 278-285. doi:10.11896/jsjkx.220500254

Abstract

PDF(2412KB) ( 318 )

References | Related Articles | Metrics

As a novel computing paradigm,mobile edge computing(MEC) provides low latency and flexible computing and communication services for mobile devices by offloading computing tasks from mobile devices to the physically proximal network edge.However,because edge servers and mobile devices often belong to different parties,the conflicts of interest between them present a great challenge for MEC systems.Therefore,it is important to design a pricing and computing offloading scheme for MEC systems with multiple edge servers and mobile devices to maximize the utility of edge servers and optimize the quality of experience for mobile devices.Considering the complex interaction between multi-edge servers and mobile devices,the multi-leader and multi-follower Stackelberg model is used to analyze the interaction between them.The edge server acts as the leader to set the price for its computing resources,and the mobile device as the follower adjusts the offloading strategy according to the pricing of the edge server.Based on the Stackelberg model,a distributed iterative algorithm based on subgradient method is proposed,which can effectively converge to Stackelberg equilibrium.Simulation results show that the proposed scheme can improve the utility of edge server and guarantee the experience quality of mobile devices.

Multi-objective Online Hybrid Traffic Scheduling Algorithm in Time-sensitive Networks

WANG Jiaxing, YANG Sijin, ZHUANG Lei, SONG Yu, YANG Xinyu

Computer Science. 2023, 50 (7): 286-292. doi:10.11896/jsjkx.220500178

Abstract

PDF(2532KB) ( 413 )

References | Related Articles | Metrics

The time-sensitive network(TSN) based on the Ethernet protocol meets various requirements such as real-time transmission and interconnection of industrial networks through different types of streams.However,when time-triggered(TT) streams,audio/video bridging(AVB) streams,and best-effort(BE) streams are transmitting in the network,it is unavoidable that the same stream competes for queues and different streams interfere with each other.Aiming at the problem that multiple traffic scheduling in TSN affects the end-to-end delay determinism,this paper proposes an improved particle swarm optimization(PSO) for online mixed-traffic analysis.The algorithm dynamically calculates paths for mixed traffic based on network conditions,and accelerates searches to meet the time limits of online computation by reducing redundant searches and constraining particle velocities to avoid particles falling into local optimizations.What’s more,the algorithm sets corresponding fitness functions for different types of traffic to reduce mutual interference between mixed traffic and queuing delay.Simulation results show that proposed algorithm can effectively improve the success rate of mixed traffic transmission in TSN network,and has stable performance and good computing efficiency.

Locating Third-party Library Functions in Obfuscated Applications

YUAN Jiangfeng, LI Haoxiang, YOU Wei, HUANG Jianjun, SHI Wenchang, LIANG Bin

Computer Science. 2023, 50 (7): 293-301. doi:10.11896/jsjkx.221100147

Abstract

PDF(2478KB) ( 248 )

References | Related Articles | Metrics

Third-party libraries are an important part of Android applications.When enforcing security enhancement or analysis based on application repackaging,it is often necessary to locate specific functions in third-party library.To this end,there is a need to map the functions of the third-party library to the disassembly code of the target application.However,many applications are obfuscated,which brings challenges to locating third-party library functions.In the disassembly code of the obfuscated application,the discriminated fingerprints are often eliminated,hence the code becomes obscure and difficult to analyze.Due to the lack of location fingerprints,it is very difficult to identify a specific function from the huge code space.So far,the existing studies only focus on identifying which third-party libraries are included in the target application rather than locating specific functions.In this paper,a method to locate the third-party functions in obfuscated applications is presented.In the first place,the obfuscator and obfuscation parameters used in the target application are identified.The source code of the third-party library is obfuscated in the same way as done for the target application.The stage is called as obfuscation alignment in this study.On this basis,we introduce some location fingerprints into the target functions with static instrumentation,and extract the structural features to identify the function location from the target application.Experiments show that the proposed method can identify the obfuscation tools and obfuscation parameters with high accuracy,and can accurately locate the third-party library functions for popular obfuscated close-source applications.

Analysis and Improvement on Identity-based Remote Data Integrity Verification Scheme

WANG Shaohui, ZHAO Zhengyu, WANG Huaqun, XIAO Fu

Computer Science. 2023, 50 (7): 302-307. doi:10.11896/jsjkx.220600067

Abstract

PDF(1856KB) ( 242 )

References | Related Articles | Metrics

Cloud storage services enable individuals or enterprises to easily maintain and manage large amounts of data at a low cost,but they cannot guarantee the integrity and privacy of outsourced data at the same time.The remote data integrity verification schemes allow users to verify the integrity of outsourced data without downloading all the data,that is,the cloud server can prove to the verifier that it is actually store the user′s data honestly.The security of an identity-based privacy preserving remote data integrity verification scheme proposed by Li et al.is analyzed.The analysis shows that the scheme is subjected to forgery attack,that is,the cloud server only needs to store a small amount of data to generate legitimate data integrity proof.Based on Li et al.’s scheme,a new identity-base remote data integrity verification scheme is proposed.The analysis shows that the new scheme can meet the security requirements of privacy and soundness,and the computational cost is basically comparable to that of Li et al.’s scheme.

Intelligent Attack Path Discovery Based on Hierarchical Reinforcement Learning

ZENG Qingwei, ZHANG Guomin, XING Changyou, SONG Lihua

Computer Science. 2023, 50 (7): 308-316. doi:10.11896/jsjkx.220500101

Abstract

PDF(2691KB) ( 429 )

References | Related Articles | Metrics

Intelligent attack path discovery is a key technology for automated penetration testing,but existing methods face the problems of exponential growth of state and action space and sparse rewards,which make the algorithm difficult to converge.To this end,an intelligent attack path discovery method(iPathD) based on hierarchical reinforcement learning is proposed.iPathD constructs the attack path discovery process as a layered Markov decision process to describe the upper-layer inter-host penetration path discovery and the lower-layer single-host internal attack path discovery,respectively.On this basis,an attack path discovery algorithm based on hierarchical reinforcement learning is proposed and implemented.Experimental results show that compared with the traditional method based on deep Q learning(DQN) and its improved algorithm,the iPathD path discovery method is faster and more effective.With the increase of the number of vulnerabilities in the host,the effect of iPathD is better,and it is suitable for large-scale network scenarios.

Browser Fingerprint Recognition Based on Improved Self-paced Ensemble Algorithm

ZHANG Desheng, CHEN Bo, ZHANG Jianhui, BU Youjun, SUN Chongxin, SUN Jia

Computer Science. 2023, 50 (7): 317-324. doi:10.11896/jsjkx.220600068

Abstract

PDF(2536KB) ( 271 )

References | Related Articles | Metrics

Browser fingerprinting technology has been used by many websites for user tracking,advertising delivery and security verification due to its stateless,cross-domain consistency and other advantages.The task of browser fingerprint recognition is a typical classification task of imbalanced data.The data imbalance exists in browser fingerprint long-term tracking task,which will lead to low accuracy of fingerprint recognition and failure of long-term tracking.An improved Self-paced Ensemble(ISPE) method is proposed to identify browser fingerprints.And the undersampling process of browser fingerprint sample and the training process of single classifier in ensemble learning are improved.Focusing on the browser fingerprint which is difficult to identify,added attention-like mechanism and self-paced factor are optimized to make the classifier pay more attention to the boundary samples which are difficult to classify in the training process,to improve the overall accuracy of browser fingerprint recognition.The results show that the F1-score of ISPE algorithm for browser fingerprint recognition reaches 95.6%,which is 16.8% higher than that of Bi-RNN algorithm.It proves that the method has excellent performance for long-term browser fingerprint tracking.

Adversarial Malware Generation Method Based on Genetic Algorithm

LI Kun, GUO Wei, ZHANG Fan, DU Jiayu, YANG Meiyue

Computer Science. 2023, 50 (7): 325-331. doi:10.11896/jsjkx.220800176

Abstract

PDF(2162KB) ( 277 )

References | Related Articles | Metrics

In recent years,with the development of Internet technology,malware has become an important method of network attack.To defend against malware attacks,deep learning techniques can be applied to malware detection.However,due to the limitations of deep learning models,malware detection models based on deep learning are vulnerable to adversarial malware,which leads to adversarial malware evading model detection.By studying the generation of adversarial malware,it can help modeldesig-ners to improve model design,improve model robustness and defense capabilities.Therefore,for the malware detection model based on grayscale image,the adversarial malware generation method based on genetic algorithm is proposed.It optimizes the perturbation by genetic algorithm,and then injects the perturbation into the malware by the obfuscation operation,so as to ensure that the generated adversarial malware samples are adversarial,executable and malicious.It is verified by experiments that the attack success rate of adversarial samples generated by the proposed method increases by 56.4% on average compared to the exis-ting work.

Data Reconstruction Attack for Vertical Graph Federated Learning

LI Rongchang, ZHENG Haibin, ZHAO Wenhong, CHEN Jinyin

Computer Science. 2023, 50 (7): 332-338. doi:10.11896/jsjkx.220900038

Abstract

PDF(1990KB) ( 264 )

References | Related Articles | Metrics

Recently,data privacy protection regulations restrict the direct exchange of raw data between different graph data ow-ners,resulting in the phenomenon of “data silos”.To solve this problem,vertical federated learning graph neural network realizes distributed training of graph data by secretly exchanging embeddings,and has been widely used in many real-world fields,such as drug discovery,user discovery,and product recommendation.However,honest participants in vertical federated learning graph neural network still have the risk of privacy leakage during training.This paper proposes a private embedding representation reconstruction attack based on the generative network,and reconstructs the private data of the participant by the output of the ge-nerative network is approximated with the confidence published from server with the norm loss function.Experimental results show that the embedding representation reconstruction attack can completely reconstruct the embedding representation of the participants on the Cora,Citeseer and Pubmed datasets,which highlights the risk of leakage of the participant embedding representation in VFL-GNN.

Fine Grained and Efficient Searchable Encryption Scheme Based on Attribute Policy Hiding inCloud Environment

ZHOU Yihua, LI Meiqi, HU Xinyu, YANG Yuguang

Computer Science. 2023, 50 (7): 339-346. doi:10.11896/jsjkx.220500238

Abstract

PDF(2209KB) ( 262 )

References | Related Articles | Metrics

Attribute based encryption provides flexible and fine-grained access control for outsourced data stored in the cloud.The traditional attribute based ciphertext policy encryption scheme(CP-ABE),whose access policy often appears in the form of plaintext,is very easy to expose users’ sensitive privacy information.In addition,due to the addition of attributes,the related calculation and storage costs in the encryption,decryption and search stages are linear with the number of attributes,and policy hiding will also increase the subsequent calculation costs.These are difficult to meet the actual needs of secure and efficient searchable encryption with privacy protection in cloud environment.To solve the above problems,a searchable encryption scheme supporting both policy hiding and constant ciphertext length is proposed.Based on the multi-valued wildcard and gate strategy,the scheme realizes the constant length of the ciphertext,and has a fixed encryption,decryption and search overhead,reducing users’ computing overhead and the storage overhead of the ciphertext in the cloud.The attributes in the access policy are completely hidden by encryption,and the bloom filter is used to judge whether the user has the relevant attributes in the access policy during the search,which not only protects users’ privacy,but also improves the computing efficiency.The scheme meets the IND-CPA safety under the assumption of q-BDHE.Security analysis and experimental results show that the scheme is safe,efficient and feasible.It is an efficient keyword search scheme,and has a good application prospect in cloud environment and Internet of Things.

Reconstructing the Right to Algorithm Explanation －－Full Algorithm Development Flow Governance and Hierarchical Classification Interpretation Framework

CONG Yingnan, WANG Zhaoyu, ZHU Jinqing

Computer Science. 2023, 50 (7): 347-354. doi:10.11896/jsjkx.220900120

Abstract

PDF(1469KB) ( 266 )

References | Related Articles | Metrics

With the rapid development of artificial intelligence,automated decision-making algorithms(ADM) have gradually entered the public domain and increasingly affected social welfare and individual interests.Meanwhile,emerging risks of ADM,such as algorithmic discrimination,algorithmic bias,and algorithm monopoly have raised the demand of governance to algorithm.Faced with information and technology asymmetry among parties involved,traditional legal resources fall short in protecting the rights of users in ADM,which justifies the right to explanation.In addition,the right to algorithm explanation,serving as an important means of algorithm governance,is conductive to making the black box of algorithm moderately transparent,correcting information asymmetry,and balancing the risk burden between the deployer and the user.It has thus become a necessity in regulating ADM deployers and safeguarding the interests of its users.Therefore,the right to explanation has become the focus in both academic and practical realms from home and abroad.However,the right to algorithm explanation in China is faced with the problem of limited eligible parties,insufficient protection scope,and inexplicit content of rights.In this regard,this paper advocates decons-tructing the right to explanation and further reconstructing it from the perspective of machine learning workflow with a hierarchical classification framework.Introducing the concept of machine learning workflow can reasonably extend the scope of the subject and object of the right,while establishing the framework of hierarchical classification can clarify the content and boundary of the right,which considers both individuality and generality of algorithms and balances the efficiency of explanation and the protection of users’ rights.In this way,all parties in ADM can be fully protected,and the development of digital economy can be empo-wered.

Water Resources Governance Mode in Watersheds Oriented to “RNAO-Ecology” Hypernetwork Complex Structure

SUO Liming, LI Jun

Computer Science. 2023, 50 (7): 355-367. doi:10.11896/jsjkx.220900134

Abstract

PDF(2201KB) ( 286 )

References | Related Articles | Metrics

For a long time,the integrity of watersheds and the fragmentation of authority have resulted in high cost of water resources governance in China’s watersheds.The transition to network governance has become a hot topic of watershed research in recent years,and a broad consensus has been formed.Compared with the three traditional schemes proposed by western scholars,the RNAO(Restricted Network Administration Organization)network structure focusing on coordination strategies is more in line with the practical needs of localization of water resources governance in the watersheds.A more general governance theory for RNAO is the direction of its theoretical development.First,this paper sorts out the triple advanced logic of the “traditional-network-hypernetwork” process of water resources governance of watershed research.Secondly,it integrates RNAO and “social-ecological” system theory,innovatively proposes an “RNAO-ecological” scale matched watershed hypernetwork governance mode,and analyzes the complex interaction mechanism of the three-layer sub-network of “organization network,behavior network,and ecological network” in detail,thus initially forming a general theoretical construction of RNAO.Finally,according to the two cases of “united river chief system” and “water resource governance of Heihe watershed”,this paper analyzes the application of RNAO and “RNAO-ecology” system in China’s watershed governance practice and gives relevant policy suggestions and possible academic topics for the future network transformation of watershed governance.

Decoupling Analysis of Network Structure Affecting Propagation Effect

CUI Yunsong, WU Ye, XU Xiaoke

Computer Science. 2023, 50 (7): 368-375. doi:10.11896/jsjkx.220900113

Abstract

PDF(3586KB) ( 234 )

References | Related Articles | Metrics

As more and more people spread information through social networks,online social networks have gradually changed the way people exchange information,so the influencing factors of information dissemination effect based on online social networks have attracted the attention of many researchers,especially the influence of network structure on the effect of information dissemination.In previous studies,most of the research has emphasized the influence of a certain network statistic on the effect of information dissemination,but the phenomenon of coupling between various network statistics is objective,and the change of a certain network statistic may lead to synchronous changes of other network statistics,which may affect the final propagation effect.In this study,a decoupled zero model framework is proposed,which decouples the coupling between different network statistics through the zero model,and then uses the SIR propagation model simulation experiment to analyze the influence of network statistics on information propagation effect without coupling,and finally uses linear threshold model simulation experiments to verify the applicability of the experimental conclusions of the SIR model in social reinforcement.Simulation experiments of the propagation model of Facebook and Twitter empirical networks show that the average shortest path of the network is the main factor affecting the speed and scope of information dissemination,and the clustering coefficient is the secondary factor affecting the scope of information dissemination.

Modeling and Simulation of Point-to-Point Propagation of False Information Based on Information Risk Perception

YU Kai, SU Tianrui

Computer Science. 2023, 50 (7): 376-385. doi:10.11896/jsjkx.220900084

Abstract

PDF(4085KB) ( 222 )

References | Related Articles | Metrics

In the study of computational social science,the propagation model inspired by infectious diseases is widely used to simulate the spread of false information.However,the traditional infectious disease model does not distinguish the differences among individuals.In the real world,the difference between individuals helps to understand how false information spreads between individuals ,which is of great significance for exploring the propagation law of false information in social networks and suppressing the spread of false information.Based on the information risk perception theory,this paper makes use of online social users’ emotion,knowledge level,trust and the number of media contacts to distinguish the differences of communication indivi-duals,and builds a more realistic point-to-point communication model of false information.In the process of communication,the differences between individuals are manifested in different communication probabilities.Individuals with high communication probabilities are more likely to be transformed into communication states.This paper uses the manually annotated Facebook data set to conduct simulation to study the propagation laws of false information.The results show that,compared with the average probability propagation system,the time span of spreading false information in the point-to-point propagation mode will be longer and the coverage will be wider.In addition,by controlling the nodes with high propagation probability,it is possible to control the propagation of false information in advance,and it has achieved better results than the methods of random control and controlling the nodes with high influence.However,increasing the control proportion of nodes in turn can not achieve better control results as expected,and the characteristics of “anti common sense” appear.