Computer Science

New Cryptographic Primitive: Definition, Model and Construction of Ratched Key Exchange

FENG Deng-guo

Computer Science. 2022, 49 (1): 1-6. doi:10.11896/jsjkx.yg20220101

Abstract

PDF(1385KB) ( 1639 )

References | Related Articles | Metrics

In the application of traditional cryptography,people always assume that the endpoints are secure and the adversary is on the communication channel.However,the prevalence of malware and system vulnerabilities makes endpoint compromise a se-rious and immediate threat.For example,it is vulnerable to various attacks such as memory content being destroyed by viruses,randomness generator being corrupted,etc.What's worse,protocol sessions usually have a long lifetime,so they need to store session-related secret information for a long time.In this situation,it becomes essential to design high-strength security protocols even in the setting where the memory contents and intermediate values of computation (including the randomness) can be exposed.Ratchet key exchange is a basic tool to solve this problem.In this paper,we overview the definition,model and construction of ratchet key exchange,including unidirectional ratcheted key exchange,sesquidirectional ratcheted key exchange and bidirectionalratcheted key exchange,and prospect the future development of ratchet key exchange.

Meta Knowledge Intelligent Systems on Resolving Logic Paradoxes

Jeffrey ZHENG

Computer Science. 2022, 49 (1): 9-16. doi:10.11896/jsjkx.210700023

Abstract

PDF(4996KB) ( 1410 )

References | Related Articles | Metrics

Professor Q.S.GAO (Chinese Science Academician) published New Fuzzy Set Theory in 2006 to explore possible solutions removing paradoxes in Fuzzy logic.In 2009,he published Foundation of Unified Linguistics from Science Press to provide bases of theoretical supports on computational multiple linguistics.The two monographs are the topmost invaluable diamonds in his creative academic activities.In memory of professor Q.S.Gao passed away for 10 years,it is my great pleasure to use new vector logic-variant construction,to describe the newest development on meta knowledge construction following advanced researches of professor Gao's legacy.Starting from vector logic,conjugate structure,meta knowledge model and other advanced mechanisms,it is a critical condition to use modern logic and mathematics to guarantee a complex system to be a consistent-dynamic one without paradoxes,to avoid if the complex system contains any logic paradox.From a classified and adjudicate viewpoint,paradoxes are divided into two categories:logic paradoxes,and semantic paradoxes.Using conjugate ring,it systematically resolves single surface property of Mobius ring to be four colored bands that support possible for this construction to resolve a series of intrinsic logicparadoxes in geometry,topology and logic.Conjugate ring provides a complete solution to resolve the Mo-bius type of paradoxes in general.Corresponding structures include many abstract systems,such as I Ching,differential geometry,geometric topology,global variation and optimization etc.Associated with resolving the Mobius type of paradoxes on topology,geometry and logic,it is natural for meta knowledge model to establish relevant key modules to support complex natural/artificial knowledge systems.Starting from classic logic,typical components are listed,such as classical logic,finite automata,Turing machine and Von Neumann architecture.Applying vector logic construction,conjugate structure and variant construction as key components with paradox-free properties,it is convenient to establish a series of architectures to support quantum Turing machine,vector machine on multiple complex functions,complicated intelligent systems,and analysis system of unified linguistics.This is an initial step for meta knowledge model to create future complicated intelligent systems.

Incorporating Language-specific Adapter into Multilingual Neural Machine Translation

LIU Jun-peng, SU Jin-song, HUANG De-gen

Computer Science. 2022, 49 (1): 17-23. doi:10.11896/jsjkx.210900005

Abstract

PDF(1989KB) ( 1079 )

References | Related Articles | Metrics

Multilingual neural machine translation (mNMT) leverages a single encoder-decoder model for translations in multiple language pairs.mNMT can encourage knowledge transfer among related languages,improve low-resource translation and enable zero-shot translation.However,the existing mNMT models are weak in modeling language diversity and perform poor zero-shot translation.To solve the above problems,we first propose a variable dimension bilingual adapter based on the existing adapter architecture.The bilingual adapters are introduced in-between each two Transformer sub-layers to extract language-pair-specific features and the language-pair-specific capacity in the encoder or the decoder can be altered by changing the inner dimension of adapters.We then propose a shared monolingual adapter to model unique features for each language.Experiments on IWSLT dataset show that the proposed model remarkably outperforms the multilingual baseline model and the monolingual adapter can improve the zero-shot translation without deteriorating the performance of multilingual translation.

Similarity-based Curriculum Learning for Multilingual Neural Machine Translation

YU Dong, XIE Wan-ying, GU Shu-hao, FENG Yang

Computer Science. 2022, 49 (1): 24-30. doi:10.11896/jsjkx.210800254

Abstract

PDF(2038KB) ( 617 )

References | Related Articles | Metrics

Multilingual neural machine translation (MNMT) with a single model has drawn more attention due to its capability to deal with multiple languages.However,the current multilingual translation paradigm does not make use of the similar features embodied in different languages,which has already been proven useful for improving the multilingual translation.Besides,the training of multilingual model is usually very time-consuming due to the huge amount of training data.To address these problems,we propose a similarity-based curriculum learning method to improve the overall performance and convergence speed.We propose two hierarchical criteria for measuring the similarity,one is for ranking different languages (inter-language) with singular vector canonical correlation analysis,and the other is for ranking different sentences in a particular language (intra-language) with cosine similarity.At the same time,the paper proposes a curriculum learning strategy that takes the loss of validation set as the curriculum replacement standard.We conduct experiments on balanced and unbalanced IWSLT multilingual data sets and Europarl corpus datasets.The results demonstrate that the proposed method outperforms strong multilingual translation systems and can achieve up to a 64% decrease in training time.

Survey of Mongolian-Chinese Neural Machine Translation

HOU Hong-xu, SUN Shuo, WU Nier

Computer Science. 2022, 49 (1): 31-40. doi:10.11896/jsjkx.210900006

Abstract

PDF(2542KB) ( 1055 )

References | Related Articles | Metrics

Machine translation is the process of using a computer to convert one language into another language.With the deep understanding of semantics,neural machine translation has become the most mainstream machine translation method at present,and it has made remarkable achievements in many translation tasks with large-scale alignment corpus,but the effect of translation tasks for some low-resource languages is still not ideal.Mongolian-Chinese machine translation is currently one of the main low-resource machine translation studies in China.The translation of Mongolian and Chinese languages is not simply the conversion between the two languages,but also the communication between the two nations,so it has attracted wide attention at home and abroad.This thesis mainly expounds the development process and research status of Mongolian-Chinese neural machine translation,and then selects the frontier methods of Mongolian-Chinese neural machine translation research in recent years,including data augmentation methods based on unsupervised lear-ning and semi-supervised learning,reinforcement learning,adversarial lear-ning,transfer-learning and neural machine translation methods assisted by pre-training models,etc.,and briefly introduce these methods.

Construction Method of Parallel Corpus for Minority Language Machine Translation

LIU Yan, XIONG De-yi

Computer Science. 2022, 49 (1): 41-46. doi:10.11896/jsjkx.210900012

Abstract

PDF(1581KB) ( 1455 )

References | Related Articles | Metrics

The training performance of neural machine translation depends heavily on the scale and quality of parallel corpus.Unlike some common languages,the construction of high-quality parallel corpora between Chinese and minority languages has been lagging.The existing minority language parallel corpora are mostly constructed by using automatic sentence alignment technology and network resources,which has many limitations such as domain and quality confined.Although high-quality parallel corpora could be constructed by manual,it lacks relevant experience and method.From the perspective of machine translation practitioners and researchers,this article introduces a cost-effective method to manually construct parallel corpus between minority languages and Chinese,including its overall goals,implementation process,engineering details,and the final result.This article tries and accumulats various experiences in the construction process,and finally forms a summary of the methods and suggestions for constructing parallel corpora from minority languages to Chinese.In the end,this paper successfully constructs 0.5 million high-quality parallel corpora from Persian to Chinese,Hindi to Chinese,and Indonesian to Chinese.The experimental results prove the quality of our constructed corpora,and it improves the performance of the minority language neural machine translation models.

Latest Development of Multilingual Speech Recognition Acoustic Model Modeling Methods

CHENG Gao-feng, YAN Yong-hong

Computer Science. 2022, 49 (1): 47-52. doi:10.11896/jsjkx.210900013

Abstract

PDF(1410KB) ( 1508 )

References | Related Articles | Metrics

With the rapid development of multimedia and communication technology,the amount of multilingual speech data on the Internet is increasing.Speech recognition technology is the core for media analysis and processing.How to quickly expand from a few major languages such as Chinese and English to more languages has become a prominent issue yet to be overcome in order to improve multilingual processing capabilities.This article summarizes the latest progress in the field of acoustic model modeling,and discusses breakthroughs needed by traditional speech recognition technology in the course of moving from single language to multi-languages.The latest end-to-end speech recognition technology was exploited to construct a keyword spotting system,and the system achieves favorable performance.The approach is detailed as follows:1)multi-lingual hierarchical and structured acoustic model modeling method;2)multilingual acoustic modeling based on language classification information;3)end-to-end keyword spotting based on frame-synchronous alignments.

Study on Keyword Search Framework Based on End-to-End Automatic Speech Recognition

YANG Run-yan, CHENG Gao-feng, LIU Jian

Computer Science. 2022, 49 (1): 53-58. doi:10.11896/jsjkx.210800269

Abstract

PDF(1586KB) ( 930 )

References | Related Articles | Metrics

In the past decade,end-to-end automatic speech recognition (ASR) frameworks have developed rapidly.End-to-end ASR has shown not only very different characteristics from traditional ASR based on hidden Markov models (HMMs),but also advanced performances.Thus,end-to-end ASR is being more and more popular and has become another major type of ASR frameworks.A keyword search (KWS) framework based on end-to-end ASR and frame-synchronous alignment is proposed for solving the problem that end-to-end ASR cannot provide accurate keyword timestamps and confidence scores,and experimental verification on a Vietnamese dataset is made.First,utterances are decoded by an end-to-end Uyghur ASR system,obtaining N-best hypotheses.Next,a dynamic programming-based alignment algorithm is implemented on each of these ASR hypotheses and per-frame phoneme probabilities,which are provided by a phoneme classifier jointly trained with the ASR model,to compute time stamps and confidence scores for each word in N-best hypotheses.Then,final KWS result is obtained by detecting keywords within N-best hypotheses and removing duplicated keyword occurrences according to time stamps and confident scores.Experimental results on a Vietnamese conversational telephone speech dataset show that the proposed KWS system achieves an F1 score of 77.6%,which is relatively 7.8% higher than the F1 score of the traditional HMM-based KWS system.The proposed system also provides reliable keyword confidence scores.

Query-by-Example with Acoustic Word Embeddings Using wav2vec Pretraining

LI Zhao-qi, LI Ta

Computer Science. 2022, 49 (1): 59-64. doi:10.11896/jsjkx.210900007

Abstract

PDF(1823KB) ( 911 )

References | Related Articles | Metrics

Query-by-Example is a popular keyword detection method in the absence of speech resources.It can build a keyword query system with excellent performance when there are few labeled voice resources and a lack of pronunciation dictionaries.In recent years,neural acoustic word embeddings has become a commonly used Query-by-Example method.In this paper,we propose to use wav2vec pre-training to optimize the neural acoustic word embeddings system,which is using bidirectional long short-term memory.On the data set extracted in SwitchBoard,the features extracted by the wav2vec model are directly used to replace the Mel frequency cepstral coefficient features,which relatively increases the system's average precision rate by 11.1% and precision recall break-even point by 10.0%.Subsequently,we tried some methods to fuse the wav2vec feature and Mel frequency cepstral coefficient feature to extract the embedding vector.The average precision rate and precision recall break-even point of the fusion method is a relative increase of 5.3% and 2.5% compared to the method using only wav2vec.

Survey of Multilingual Question Answering

LIU Chuang, XIONG De-yi

Computer Science. 2022, 49 (1): 65-72. doi:10.11896/jsjkx.210900003

Abstract

PDF(1925KB) ( 1012 )

References | Related Articles | Metrics

Multilingual question answering is one of the research hotspots in the field of natural language processing,which aims to enable the model to return a correct answer based on understanding of the given questions and texts in different languages.With the rapid development of machine translation technology and the wide application of multilingual pre-training technology in the field of natural language processing,multilingual question answering has also achieved a relatively rapid development.This paper first systematically reviews the current work of multilingual question answering methods,and divides them into feature-based methods,translation-based methods,pre-training-based methods and dual encoding-based methods,and introduces the use and characteristics of each method respectively.Meanwhile,it also discusses the current work related to multilingual question answe-ring tasks,and divides them into text-based and multi-modal-based tasks and gives the basic definition of each one.Moreover,this paper summarizes the dataset statistics,evaluation metrics and multilingual question answering methods involved in these tasks.Finally,it proposes the future research prospect of multilingual question answering.

Improving Low-resource Dependency Parsing Using Multi-strategy Data Augmentation

XIAN Yan-tuan, GAO Fan-ya, XIANG Yan, YU Zheng-tao, WANG Jian

Computer Science. 2022, 49 (1): 73-79. doi:10.11896/jsjkx.210900036

Abstract

PDF(2164KB) ( 665 )

References | Related Articles | Metrics

Dependency parsing aims to identify syntactic dependencies between words in a sentence.Dependency parsing can provide syntactic features and improve model performance for tasks such as information extraction,automatic question answering and machine translation.The training data size has an significant impact on the performance of the dependency parsing model.The lack of training data will cause serious unknown word problems and model over-fitting problems.This paper proposes various data augment strategies for the problem of low-resource dependency parsing.The proposed method effectively expands the training data by synonym substitution and alleviates the unknown words problem.The data augment strategies of multiple Mixups effectively alleviate the model overfitting problem and improve the generalization ability of the model.Experimental results on the universal dependencies treebanks(UD treebanks) dataset show that the proposed methods effectively improve the performance of Thai,Vietnamese and English dependency parsing under small-scale training corpus conditions.

Imbalanced Data Classification:A Survey and Experiments in Medical Domain

JIANG Hao-chen, WEI Zi-qi, LIU Lin, CHEN Jun

Computer Science. 2022, 49 (1): 80-88. doi:10.11896/jsjkx.210200124

Abstract

PDF(2115KB) ( 3006 )

References | Related Articles | Metrics

In recent years,AI technology has been widely adopted in many application domains,amongst which,intelligent medical applications such as clinical decision support systems have attracted much attention.However,since the current wave of AI applications are based on predictive models crystalized from historical data,the feature and quality of data will affect AI applications' performance directly.Medical data are inherently imbalanced as rare disease cases are always the scarce in existing case archives,while considered more important.The “data imbalance problem” is still considered a difficult research problem in machine lear-ning.This paper conducts a literature review on the research efforts targeting at techniques to handle “imbalanced data” in gene-ral as well as the ones in intelligent medical area.We then use research publications from the SIGKDD conference dedicated to knowledge discovery and data mining as a sample pool,to find people's preferred approach to address “imbalanced data” problem in a given domain.Finally,based on approaches,we identify from the survey,and conduct experiments on two typical medical predictive model learning scenarios,to validate the know-how we acquired in this study.

Community Detection Algorithm for Dynamic Academic Network

PU Shi, ZHAO Wei-dong

Computer Science. 2022, 49 (1): 89-94. doi:10.11896/jsjkx.210100023

Abstract

PDF(1701KB) ( 2636 )

References | Related Articles | Metrics

Academic network is a kind of dynamic heterogeneous information network.Community detection on the academic network can dig out the communities of academic subjects and discover the insights contained in the community structure.The exis-ting community detection algorithms ignore the dynamics of the academic network and the special relationship between academic subjects and do not optimize the closeness of the academic community and the relationship between academic communities.This paper proposes a community detection algorithm called DANE-CD based on dynamic academic network representation learning.Firstly,an autoencoder is adopted to represent the academic subject in the academic network.Secondly,the clustering optimization based on modularity and team faultlines is innovatively integrated into the representation learning process.Finally,a dynamic academic network representation model is constructed based on the stacked autoencoder,together with the completion of community detection in the dynamic academic network.Extensive experiments on two real-world academic datasets(DBLP and HEP-TH) demonstrate that DANE-CD is superior to the baseline methods and can detect the academic communities effectively.

Distributed Distance Join Algorithm for Massive Spatial Data

WANG Ru-bin, LI Rui-yuan, HE Hua-jun, LIU Tong, LI Tian-rui

Computer Science. 2022, 49 (1): 95-100. doi:10.11896/jsjkx.210100060

Abstract

PDF(2162KB) ( 1039 )

References | Related Articles | Metrics

Spatial distance join is one of the most common operations for spatial data analysis,which has various application scenarios.Existing distributed methods face the problems of too large space,high data skew,and slow self-join.To this end,this paper proposes a novel distributed distance join algorithm,i.e.,JUST-Join,for massive spatial data.First,JUST-Join regards only the necessary space as the global domain,which can filter invalid data out,reducing the overhead of unnecessary data transmission and computation.Second,we consider both the spatial distributions of the two datasets,which relieves the data skew issue.Third,for the spatial self-join,we adopt plane sweep method to further improve the efficiency.We implement JUST-Join algorithm based on Spark,and conduct extensive experiments using real datasets.The experimental results show that JUST-Join is superior to the state-of-the-art distributed spatial analysis systems in terms both of efficiency and scalability.

DeepFM and Convolutional Neural Networks Ensembles for Multimodal Rumor Detection

CHEN Zhi-yi, SUI Jie

Computer Science. 2022, 49 (1): 101-107. doi:10.11896/jsjkx.201200007

Abstract

PDF(2092KB) ( 2847 )

References | Related Articles | Metrics

With the increasing popularity of social media represented by Weibo,rumors spread rapidly through social media,which is more likely to cause serious consequences.The problem of automatic rumor detection has attracted widespread attention from academic and industrial circles at home and abroad.We have noticed that more and more users use pictures to post Weibo,not just text.Weibo usually consists of text,images and social context.Therefore,a multi-modal network rumor detection method DCNN based on deep neural network for the text content,image and user attribute information of the accompanying text is proposed.This method consists of a multi-modal feature extractor and a rumor detector.The multi-modal feature extractor is divided into three parts:a text feature extractor based on TextCNN,a picture feature extractor based on VGG-19,and a user social feature extractor based on DeepFM algorithm.These three parts learn feature representations on different modalities of Weibo to form re-parameterized multi-modal features,which are fused as input to the rumor detector classification detection.This algorithm has carried out a large number of experiments on the Weibo data set,and the experimental results show that the recognition accuracy of DCNN algorithm is improved from 78.1% to 80.3%,which verifies the feasibility and effectiveness of DCNN algorithm and feature interaction method for social characteristics.

Multivariate Regression Forest for Categorical Attribute Data

LIU Zhen-yu, SONG Xiao-ying

Computer Science. 2022, 49 (1): 108-114. doi:10.11896/jsjkx.201200189

Abstract

PDF(1623KB) ( 792 )

References | Related Articles | Metrics

As categorical attributes cannot be utilized directly in some regression models like the linear regression,SVR and most multivariate regression trees,a multivariate split method dealing with multiple types of data is prompted in this paper.We define the centers of the sample sets on the categorical attributes and the distances from the samples to the centers in order that thecate-gorical attributes can also participate in the clustering process like the numerical attributes.Then a reasonable ensemble scheme is selected for the decision trees generated by the method to get the ensemble called cluster regression forest(CRF).Finally,we use CRF and other 9 regression models to compare regression mean absolute error (MAE) and root mean square error (RMSE) on 12 UCI public data sets.The experimental results show that CRF has the best performance among the 10 regression models.

Convolutional Sequential Recommendation with Temporal Feature and User Preference

CHEN Jin-peng, HU Ha-lei, ZHANG Fan, CAO Yuan, SUN Peng-fei

Computer Science. 2022, 49 (1): 115-120. doi:10.11896/jsjkx.201200192

Abstract

PDF(2002KB) ( 744 )

References | Related Articles | Metrics

At present,recommendation system has been widely used in our life,which greatly facilitates people's life.The traditional recommendation method mainly analyzes the interaction between users and items and considers the history of users and items,and only obtains the user's preference for items in the past.The sequential recommendation system,by analyzing the interaction sequence of items in the recent period of time and considering the relevance between the user's previous and subsequent behaviors,can obtain user's preference for items in short term.It emphasizes the short-term connection between user and item,while ignoring the relationship between the attributes of the item.Aiming at the above problems,this paper presents a convolutional embedding recommendation with time and user preference (CERTU) model.This model can analyze the relations between items.It can obtain dynamic changes in user preferences.The model also considers the influence of individual item and multiple items to the next item.Experiments show that the performance of CERTU model is better than that of the current baseline method.

Study on Density Parameter and Center-Replacement Combined K-means and New Clustering Validity Index

ZHANG Ya-di, SUN Yue, LIU Feng, ZHU Er-zhou

Computer Science. 2022, 49 (1): 121-132. doi:10.11896/jsjkx.201100148

Abstract

PDF(2786KB) ( 516 )

References | Related Articles | Metrics

As a classical data mining technique,clustering is widely used in fields as pattern recognition,machine learning,artificial intelligence,and so on.By effective clustering analysis,the underlying structures of datasets can be identified.As a commonly used partitional clustering algorithm,K-means is simple of implementation and efficient on classifying large scale datasets.However,due to the influence of the convergence rule,the traditional K-means is still suffering problems as sensitive to the initial clustering centers,cannot properly process non-convex distributed datasets and datasets with outliers.This paper proposes the DC-Kmeans (density parameter and center replacement K-means),an improved K-means algorithm based on the density parameter and center replacement.Due to the gradually selecting of initial clustering centers and continuously update imprecision old centers,the DC-Kmeans is more accurate than the traditional K-means.Two novel methods are also proposed for optimally clustering:1)a novel clustering validity index (CVI),SCVI (Sum of the inner-cluster compactness and the inter-cluster separateness based CVI),is proposed to evaluate the results of the DC-Kmeans;2)a new algorithm,OCNS (optimal clustering number determination based on SCVI),is designed to determine the optimal clustering numbers for different datasets.Experimental results demonstrate that the proposed clustering method is effective for many kinds of datasets.

Generative Adversarial Network and Meta-path Based Heterogeneous Network Representation Learning

JIANG Zong-li, FAN Ke, ZHANG Jin-li

Computer Science. 2022, 49 (1): 133-139. doi:10.11896/jsjkx.201000179

Abstract

PDF(2247KB) ( 706 )

References | Related Articles | Metrics

Most of the information works in real world are heterogeneous information networks (HIN).Network representation methods aiming to represent node data in low dimensional space have been widely used to analyze heterogeneous information networks,so as to effectively integrate rich semantic information and structural information in heterogeneous networks.However,the existing heterogeneous networks representation methods usually use negative sampling to select nodes randomly from the network,and the heterogeneity learning ability of nodes and edges is insufficient.Inspired by the generative adversarial networks (GAN) and meta-path,we propose a new framework,which is improved by weighted meta-path based sampling strategy.The samples can better reflect the direct and indirect relationship between nodes and enhance the semantic association of samples.In the process of generation and confrontation,the model fully considers the heterogeneity of nodes and edges,and has the ability of relationship perception,so as to realize the representation learning of heterogeneous information networks.The experimental results show that,compared with the current representation algorithms,the representation vectors learned by the model have better performance in classification and link prediction experiments.

Electricity Theft Detection Based on Multi-head Attention Mechanism

XIAO Ding, ZHANG Yu-fan, JI Hou-ye

Computer Science. 2022, 49 (1): 140-145. doi:10.11896/jsjkx.210100177

Abstract

PDF(1880KB) ( 891 )

References | Related Articles | Metrics

Electricity theft causes significant damage to social and economic development.How to detect malicious electricity theft based on power big data has been widely concerned by academia and industry.Aiming at the problems of traditional methods relying on manual features,insufficient behavior sequence representation,poor detection accuracy,etc.,this paper proposes an electricity theft detection model based on multi-head attention mechanism (ETD-MHA).The bidirectional gated recurrent unit is used to fully capture the time features of the electricity consumption behavior sequence,and the distinction of key features is gradually enhanced in the multi-head attention mechanism,and finally,the learning effect is improved by deepening the networks.Extended experiments are conducted on the smart meter datasets of Ireland and China State Grid.The results show that the proposed method achieves better performance compared with the linear regression (LR),support vector machine (SVM),random forest (RF),and other traditional algorithms.For example,the AUC value of the proposed model is improved by up to 34.6%compared to the LR algorithm.

Locality and Consistency Based Sequential Ensemble Method for Outlier Detection

LIU Yi, MAO Ying-chi, CHENG Yang-kun, GAO Jian, WANG Long-bao

Computer Science. 2022, 49 (1): 146-152. doi:10.11896/jsjkx.201000156

Abstract

PDF(2104KB) ( 604 )

References | Related Articles | Metrics

Outlier detection has been widely used in many fields,such as network intrusion detection,credit card fraud detection,etc.The increase in data dimensions leads to many irrelevant and redundant features,which will obscure the relevant features and result in false positive results.Due to the sparseness and distance aggregation effects of high-dimensional data,the traditional outlier detection algorithms based on density and distance are no longer applicable.Most of the outlier detection research based on machine learning focuses on a single model,which has certain deficiencies in anti-overfitting ability.The ensemble learning model has good generalization ability,and in actual application shows better prediction accuracy than the single model.This paper proposes an outlier detection sequence integration method LCSE based on neighborhood consistency (locality and consistency based sequential ensemble method for outlier detection).Firstly,it constructs a basic model of outlier detection based on diversity,secondly,selects the abnormal candidate points according to the global integration consistency,and finally considers the local neighborhood correlation of the data to select and combine the basic model results.Experiments verify that LCSE has an average outlier detection accuracy increase of 20.7% compared with traditional methods.Compared with the ensemble methods LSCP_AOM and iForest,the performance is increased by 3.6% on average.Therefore,it is better than other ensemble methods and neural network methods.

Aided Disease Diagnosis Method for EMR Semantic Analysis

FAN Hong-jie, LI Xue-dong, YE Song-tao

Computer Science. 2022, 49 (1): 153-158. doi:10.11896/jsjkx.201100125

Abstract

PDF(1762KB) ( 669 )

References | Related Articles | Metrics

Aiming at solving the problem of auxiliary disease diagnosis for electronic medical record,the word vector and text discrimination method are applied to the semantic text analysis task.Concretely,the pre-training language model is used as the semantic representation of characters,so as to accurately express the text features.After extracting N-ary features from convolutional neural network,the capsule unit is used to cluster the features,so as to better capture the high-level semantic text features and reduce the demand for data.It is found that the combination model based on ERNIE+CNN+Capsule achieves high accuracy on the real EMR.In addition,inspired by the image style transfer,a style conversion model from EMR text to disease self-report text is trained.Based on the style conversion model,non-parallel data are used to add confrontation ideas and confusion evaluation indexes,which can effectively alleviate the problem of inconsistent distribution of training data and test data.Finally,compared with ALBERT_tiny,BERT and other models,the proposed model gets 86.89% F1 value in the EMR,which is improved by1.36%~3.68%,and 94.95% F1 value in the generalization.Experiments show that the proposed model can effectively adapt to the auxiliary disease diagnosis on the premise of ensuring high accuracy.

Expert Recommendation Algorithm for Enterprise Engineering Problems

MA Jian-hong, ZHANG Tong

Computer Science. 2022, 49 (1): 159-165. doi:10.11896/jsjkx.201200227

Abstract

PDF(2130KB) ( 545 )

References | Related Articles | Metrics

Enterprises often encounter various engineering problems in the production line,which need the help of experts to be effectively solved.The current academic recommendation system could not deeply explore the underlying knowledge connection between the problem and the solution,or could not recommend suitable experts for engineering problems.The systematic research on how to recommend experts for the enterprise engineering problems to be solved is as follows.1)The influence of experts is calculated based on the expert co-author network,and the partial order information between co-authors is formed by combining the author order information.A topic model integrating co-author partial order information is proposed:APO-ACT model,which makes the author-conference-topic (ACT) model mine core experts better and more suitable for recommendation systems.2)Problem knowledge model can mine the underlying knowledge connection between the problem and the solution.Based on the database of enterprise innovation methods problems,an expert recommendation algorithm for the text description of enterprise engineering problems to be solved combining theory,technology and experience is proposed.Experiments show that the APO-ACT-based recommendation method can better mine core experts while ensuring the recommendation accuracy,which is superior to content-based recommendation and ACT model-based recommendation.

Mining Spatial co-location Patterns with Star High Influence

MA Dong, LI Xin-yuan, CHEN Hong-mei, XIAO Qing

Computer Science. 2022, 49 (1): 166-174. doi:10.11896/jsjkx.201000186

Abstract

PDF(3709KB) ( 486 )

References | Related Articles | Metrics

The spatial co-location pattern is a group of spatial features whose instances are frequently collocated in the spatial neighborhood.Traditional spatial co-location pattern mining methods usually assume that the spatial instances are independent each other,and use participation index (PI) to measure the patterns.They don't consider the influence of different features or different instances of the same feature so that the mining results are often lack of relevance and interpretability.This paper proposes the spatial co-location pattern with star high influence which has influence in the neighborhood,and its mining method.Firstly,this paper defines two indicators to measure the influence of the pattern:influence participation index (IPI) and influence occupancy index (IOI).Secondly,a basic algorithm and pruning strategies for mining co-location patterns with star high influence are proposed.Finally,the experimental results on real and synthetic data sets show that the proposed method can discover the strong relevant co-location patterns.

Dynamic Low-sampling Ambient Occlusion Real-time Ray Tracing for Molecular Rendering

LI Jia-zhen, JI Qing-ge

Computer Science. 2022, 49 (1): 175-180. doi:10.11896/jsjkx.210200042

Abstract

PDF(2726KB) ( 561 )

References | Related Articles | Metrics

High-quality rendering in molecular visualization are particularly important for researchers to observe the structure of biomolecules.The rasterization rendering effect commonly used by mainstream molecular visualization tools is not good.Advanced ray tracing rendering technology can achieve high-quality rendering effects,however,molecular rendering methods that support ray tracing in the current tools have various problems such as platform limitations,insufficient real-performance,and poor rende-ring quality.In this paper,a dynamic low-sampling ambient occlusion real-time ray tracing for molecular rendering is proposed.A simple reprojection method for ray tracing is proposed to implement temporal denoising of low-sampling ambient occlusion in dynamic.We also propose a shadow rays packet strategy to improve the parallelism of calculation when the ray traverses the scene.Experimental results show that our method can achieve interactive rendering performance on PC,and compared with the advanced VMD-OSPRay method on the TH-2 supercomputer,our method achieves performance acceleration of 1.40 to 1.64 times,and improves the serious noise problem of dynamic images.

Multi-scale Gated Graph Convolutional Network for Skeleton-based Action Recognition

GAN Chuang, WU Gui-xing, ZHAN Qing-yuan, WANG Peng-kun, PENG Zhi-lei

Computer Science. 2022, 49 (1): 181-186. doi:10.11896/jsjkx.201100164

Abstract

PDF(2636KB) ( 630 )

References | Related Articles | Metrics

Skeleton-based human action recognition is attracting more attention in computer vision.Recently,graph convolutional networks(GCNs),which is powerful to model non-Euclidean structure data,have obtained promising performance and enable a new paradigm for action recognition.Existing approaches mostly model the spatial dependency with emphasis mechanism since the huge pre-defined graph contains large quantities of noise.However,simply emphasizing subsets is not optimal for reflecting the dynamic underlying correlations between vertexes in a global manner.Furthermore,these methods are ineffective to capture the temporal dependencies as the CNNs or RNNs are not capable to model the intricate multi-range temporal relations.To address these issues,a multi-scale gated graph convolutional network (MSG-GCN) is proposed for skeleton-based action recognition.Specifically,a gated temporal convolution module (G-TCM) is presented to capture the consecutive short-term and interval long-term dependencies between vertexes in the temporal domain.Besides,a multi-dimensional attention module for spatial,temporal,and channel,which enhances the expressiveness of spatial graph,is integrated into GCNs with negligible overheads.Extensive experiments on two large-scale benchmark datasets,NTU-RGB+D and Kinetics,demonstrate that our approach outperforms the state-of-the-art baselines.

Low-light Image Enhancement Model with Low Rank Approximation

WANG Yi-han, HAO Shi-jie, HAN Xu, HONG Ri-chang

Computer Science. 2022, 49 (1): 187-193. doi:10.11896/jsjkx.210600090

Abstract

PDF(2534KB) ( 866 )

References | Related Articles | Metrics

Due to the influence of low lightness,the images acquired at dim or backlight conditions tend to have poor visual quality.Retinex-based low-light enhancement models are effective in improving the scene lightness,but they are often limited in hand-ling the over-boosted image noise hidden in dark regions.To solve this issue,we propose a Retinex-based low-light enhancement model incorporating the low-rank matrix approximation.First,the input image is decomposed into an illumination layer I and a reflectance layer R according to the Retinex assumption.During this process,the image noise in R is suppressed via low-rank-based approximation.Then,aiming to preserve the image details in the bright regions and suppress the noise in the dark regions simultaneously,a post-fusion under the guidance of I is introduced.In experiments,qualitative and quantitative comparisons with other low-light enhancement models demonstrate the effectiveness of our method.

Adaptive Bitrate Streaming for Energy-Efficiency Mobile Augmented Reality

CHEN Le, GAO Ling, REN Jie, DANG Xin, WANG Yi-hao, CAO Rui, ZHENG Jie, WANG Hai

Computer Science. 2022, 49 (1): 194-203. doi:10.11896/jsjkx.201100107

Abstract

PDF(3997KB) ( 717 )

References | Related Articles | Metrics

With the development of the mobile augmented reality (MAR),users have higher requirements on video quality and response time on it.MAR applications offload computation-intensive tasks to the cloud or edge servers for processing.In order to provide users with high-quality rendering services,MAR needs to download massive amounts of data from cloud or edge servers.Due to the instability of network condition and the limitation of network bandwidth,data transmission will extend MAR application response time,which increases the energy consumption,and seriously affects the user experience.This paper proposes a bit-rate adaptive model based on gradient boosting regression (GBR).The model considers the different needs of users in different network conditions,analyzes the features of the 200 popular videos,finds the connection between the video features and the user requirements,and provides appropriate video bitrate configuration according to different needs,thus to achieve the goal of maintaining experience,reducing latency and saving energy.The results show that compared with the original rendered 1080p video,the proposed bitrate adaptive model can save 58% downloading time latency(19.13 ms) per frame while maintaining the user's viewing experience as much as possible.

Crack U-Net:Towards High Quality Pavement Crack Detection

ZHU Yi-fan, WANG Hai-tao, LI Ke, WU He-jun

Computer Science. 2022, 49 (1): 204-211. doi:10.11896/jsjkx.210100128

Abstract

PDF(3818KB) ( 1073 )

References | Related Articles | Metrics

Pavement cracks constitute a major potential threat to driving safety.Previous manual detection methods are highly subjective and inefficient.Current computer vision methods have limited applications in crack detection.Existing models have poor generalization capabilities and limited detection effects.To address this problem,a dense network structure of pavement crack detection,called Crack U-Net,is proposed to improve the model generalization capabilities and detection accuracy.Firstly,the dense connection structure of Crack U-Net adopts the network design from the encoder-decoder backbone network U-Net.Similar to the encoder-decoder backbone network,this structure of Crack U-Net is able to improve the utilization of feature information and to enhance the robustness of the model,as well.Secondly,the Crack U-block composed of residual blocks and mini-U is proposed as the basic convolution module of the network,which can extract more abundant crack features compared with the traditional dou-ble-layer convolution layer.Finally,dilated convolution is used in the middle layer of up sampling and down sampling in the network to fully capture the crack features,which is at the edge if the image.Crack U-Net runs on public fracture dataset and produces a series of experimental results.The experimental results show that the AIU value of this method on the dataset is 2.2% higher than the previous method,and it is better than the existing fracture segmentation accuracy and generalization.The experimental results also show that Crack U-Net model can be pruned,and the pruned model is suitable for loading to mobile devices for road crack detection.

Dual-stream Reconstruction Network for Multi-label and Few-shot Learning

FANG Zhong-li, WANG Zhe, CHI Zi-qiu

Computer Science. 2022, 49 (1): 212-218. doi:10.11896/jsjkx.201100143

Abstract

PDF(2244KB) ( 759 )

References | Related Articles | Metrics

The multi-label image classification problem is one of the most important problems in the field of computer vision,which needs to predict and output all the labels in an image.However,the number of labels to be classified in an image is often more than one,and the changeable size,posture,and position of objects in the image will increase the difficulty of classification.Therefore,how to effectively improve the accurate expression ability of image features is an urgent problem to be solved.In response to the above-mentioned problem,a novel dual-stream reconstruction network is proposed to extract features from images.Specifically,the model first proposes a dual-stream attention network to extract features based on channel information and spatial information,and uses feature stitching to make image features have both channel detail information and spatial detail information.Secondly,a reconstruction loss function is introduced to constrain the features of the dual-stream network,forcing the above two divergent features to have the same feature expression ability,thereby promoting the extracted dual-stream features to approach the ground-truth features.Experimental results on multi-label image datasets based on VOC 2007 and MS COCO show that the proposed dual-stream reconstruction network can accurately and effectively extract salient features and produce better classification accuracy.At the same time,in view of the sparse effect of reconstruction loss on model features,the proposed method is also applied to few-shot learning.The experimental results show thatthe proposed model also has good classification accuracy for few-shot learning.

Image-Text Sentiment Analysis Model Based on Visual Aspect Attention

YUAN Jing-ling, DING Yuan-yuan, SHENG De-ming, LI Lin

Computer Science. 2022, 49 (1): 219-224. doi:10.11896/jsjkx.201000074

Abstract

PDF(2378KB) ( 878 )

References | Related Articles | Metrics

Social network has become an integral part of our daily life.Sentiment analysis of social media information is helpful to understand people's views,attitudes and emotions on social networking sites.Traditional sentiment analysis mainly relies on text.With the rise of smart phones,information on the network is gradually diversified,including not only text,but also images.It is found that,in many cases,images can enhance the text rather than express emotions independently.We propose a novel image text sentiment analysis model (LSTM-VistaNet).Specifically,this model does not take the picture information as the direct input,but uses the VGG16 network to extract the image features,and then generates the visual aspect attention,and gives the core sentences in the document a higher weight,and get a document representation based on the visual aspect attention.In addition,we use the LSTM network to extract the text sentiment and get the document representation based on text only.Finally,we fuse the two groups of classification results to obtain the final classification label.On the Yelp restaurant reviews data set,our model achieves an accuracy of 62.08%,which is 18.92% higher than BiGRU-mVGG,which verifies the effectiveness of using visual information as aspect attention assisted text for emotion classification;It is 0.32% higher than VistaNet model,which proves that LSTM model can effectively make up for the defect that images in VistaNet model cannot completely cover text.

Interior Human Action Recognition Method Based on Prior Knowledge of Scene

LIU Xin, YUAN Jia-bin, WANG Tian-xing

Computer Science. 2022, 49 (1): 225-232. doi:10.11896/jsjkx.201100185

Abstract

PDF(2920KB) ( 908 )

References | Related Articles | Metrics

Currently,the recognition technology targeted at human action in an interior scene is widely used in video content understanding,home-based care,medical care and other fields,and existing researches pay more heed to the modelling of human action,while ignoring the connection between interior scene and human action in videos.With a view to making full use of the relevance between the scene information and the human motion,this paper studies the recognition approaches for human action in an interior scene based on scene-prior knowledge.Yet,the paper proposes scene-prior knowledge inflated 3D ConvNet(SPI3D).Firstly,the ResNet152 network is adopted to extract scene features for scene classification.Then,based on the results,combined with scene-prior knowledge,this paper introduces quantified scene prior knowledge,optimizes the overall objective function by constraining the weights.Additionally,aiming at the problem that most of the existing data sets focus on the characteristics of human action,whereas the scene information remains complex and plain,an interior scene-action database(SADB) is established.It is shown in experimental results,on the SADB,the recognition accuracy rate of SPI3D reaches 87.9%,6% higher than the recognition accuracy of I3D directly.It can be seen that the modelling for the recognition on human action in interior scene is featured by better performance after introducing the prior knowledge of the scene.

Image Stream From Paragraph Method Based on Scene Graph

ZHANG Wei-qi, TANG Yi-feng, LI Lin-yan, HU Fu-yuan

Computer Science. 2022, 49 (1): 233-240. doi:10.11896/jsjkx.201100207

Abstract

PDF(3031KB) ( 790 )

References | Related Articles | Metrics

The task of generating sequence images from paragraphs by generating confrontation networks can already generate higher quality images.However,when the input text involves multiple objects and relationships,the context information of the text sequence is difficult to extract,the object layout of the generated image is prone to confusion,and the generated object details are insufficient.To solve this problem,this paper proposes a method of generating sequence images based on scene graphs based on StoryGAN.First,the paragraph is converted into multiple scene graphs through graph convolution,each scene graph contains the object and relationship information of the corresponding text.Then,the bounding box and segmentation mask of the object are predicted to calculate the scene layout.Finally,according to the scene layout and the context information,a sequence of images more in line with the object and its relationship is generated.Tests on CLEVR-SV and CoDraw-SV data sets show that the me-thod in this paper can generate 64×64-pixel sequence images containing multiple objects and their relationships.Experimental results show that on the CLEVR-SV data set,the SSIM and FID of this method are improved by 1.34% and 9.49% respectively than StoryGAN.On the CoDraw-SV data set,the ACC of this method is 7.40% higher than that of StoryGAN.The proposed method improves the rationality of the layout of the generated scene,not only can generate an image sequence containing multiple objects and relationships,but also the generated image has higher quality and clearer details.

Survey on Automatic Tuning of Compilers by Machine Learning

CHI Hao-yu, CHEN Chang-bo

Computer Science. 2022, 49 (1): 241-251. doi:10.11896/jsjkx.210100113

Abstract

PDF(1692KB) ( 1637 )

References | Related Articles | Metrics

Modern compilers offer many optimization options.It is a complex problem to choose which parameter values,which combination of options and in which order to apply these options.Among them,the optimization phase ordering is the most dif-ficult one.With the improvement of traditional methods (iterative compilation combined with heuristic optimization search) and the emergence of new technologies (machine learning),it is possible to build a relatively efficient and intelligent compiler automatic tuning framework.This paper summarizes the research ideas and application methods of predecessors by investigating the related research in the past decades.Firstly,the development of compiler automatic tuning is introduced,including early manual methods,cost function driven methods,iterative compilation and machine learning-based prediction methods.Then,this paper focuses on the direct prediction based on machine learning and the automatic optimization method of iterative compilation driven by machine learning.Last but not least,several successful frameworks and some recent research results are listed.Current challenges and some key future research directions are also pointed out.

Self-attention-based BGRU and CNN for Sentiment Analysis

HU Yan-li, TONG Tan-qian, ZHANG Xiao-yu, PENG Juan

Computer Science. 2022, 49 (1): 252-258. doi:10.11896/jsjkx.210600063

Abstract

PDF(1915KB) ( 1172 )

References | Related Articles | Metrics

Text sentiment analysis is a hot field in natural language processing.In recent years,Chinese text sentiment analysis methods have been widely investigated.Most of the recurrent neural network and convolutional neural network models based on word vectors have insufficient ability to extract and retain text features.In this paper,a Chinese sentiment polarity analysis model combining bi-directional GRU (BGRU) and multi-scale CNN is proposed.First,BGRU is utilized to extract text serialization features filtered with attention mechanism.Then the convolution neural network with distinct convolution kernels is applied to attention mechanism to adjust the dynamic weights.The text is acquired by the Softmax emotional polarity.Experiments indicates that our model outperforms the state-of-the-art methods on Chinese datasets.The accuracy of sentiment classification is 92.94% on the online_shopping_10_cats dataset of ChineseNLPcorpus,and 92.75% on the hotel review dataset compiled by Tan Songbo of Chinese Academy of Sciences,which is significantly improved compared with the current mainstream methods.

Logical Reasoning Based on DNA Strand Displacement

WU Li-bo, HUANG Yu-fang

Computer Science. 2022, 49 (1): 259-263. doi:10.11896/jsjkx.210200131

Abstract

PDF(2708KB) ( 581 )

References | Related Articles | Metrics

A computational model for logical reasoning is proposed based on DNA strand displacement reactions.Firstly,this way does not rely on fluorescent labeling or other DNA experimental technique.The corresponding relationship between the concentration change of DNA strand and the value of Boolean logic signal is realized by building up a special 0-1 function and using less DNA reaction strands and strand displacement reactions as possible as we can.Then the calculation models of basic logical operations “and” “or” “not” are designed based on DNA strand displacement.Furthermore,the basic logical operations can be combined arbitrarily in use of the cascading property of DNA strand displacement,so as to solve different logical reasoning problems.Finally,the feasible solution of satisfiability problem,which is a special logical reasoning problem,is implemented through the simulation.All the DNA strand displacement reaction processes and the concentration changes of related DNA chains can be simulated by the Visual DSD software.

Disease Genes Recognition Based on Information Propagation

LI Jia-wen, GUO Bing-hui, YANG Xiao-bo, ZHENG Zhi-ming

Computer Science. 2022, 49 (1): 264-270. doi:10.11896/jsjkx.201100129

Abstract

PDF(2785KB) ( 609 )

References | Related Articles | Metrics

Genetic research in the field of life science and medicine occupies an important position,while disease genes are one of its key focuses.Accurate identification of disease-causing genes can reveal the pathogenesis of diseases at the molecular level,and provide strong support for the prevention,diagnosis,treatment and other medical stages of diseases.The key to accurately identifying disease-causing genes is to give a measure of similarity between genes.This paper uses complex networks to model biological systems and proposes a dissipative random walk model with multiple restarts to measure the degree of functional similarity between genes.Firstly,a human gene-gene interaction network is constructed based on the human gene interaction datasets on NCBI.Experiments are then carried out on KEGG's disease-gene association dataset to identify known disease-causing genes.Compared with the three existing models of SP,RWR and PRINCE,DRWMR accurately predicts 156 of 581 diseases while the remaining models predict 121.3 correctly on average.The average prediction score of DRWMR is 9.46% higher.Finally,the potential disease genes of asthma,hemophilia and PEHO syndrome are predicted and the candidate genes are found guilty for the pathologies in the literature or biological database.

Unsupervised Domain Adaptation Based on Style Aware

NING Qiu-yi, SHI Xiao-jing, DUAN Xiang-yu, ZHANG Min

Computer Science. 2022, 49 (1): 271-278. doi:10.11896/jsjkx.201200094

Abstract

PDF(2032KB) ( 576 )

References | Related Articles | Metrics

In recent years,neural machine translation has made significant progress in translation quality,but it relies on parallel bilingual sentence pairs heavily during the training process.However,parallel resources are scarce for the e-commerce domain,in addition,cultural differences lead to stylistic differences in product information expression.To solve these two problems,a style-aware unsupervised domain adaptation algorithm is proposed,which makes full use of e-commerce monolingual data in the mutual training method,while introducing quasi knowledge distillation approach to deal with style differences.We construct non-parallel bilingual corpus by obtaining e-commerce product data information,and then carry out experiments based on the aforementioned corpus and Chinese and English news parallel corpus.The results show that the algorithm significantly improves translation qua-lity compared to various unsupervised domain adaptation methods,improves about 5 BLEU points compared with the strongest baseline system.In addition,the algorithm is further extended to Ted,Law and Medical OPUS data,all of which achieve better translation results.

Efficient Computation of Intervention in Causal Bayesian Networks

LI Chao, QIN Biao

Computer Science. 2022, 49 (1): 279-284. doi:10.11896/jsjkx.210300028

Abstract

PDF(1532KB) ( 547 )

References | Related Articles | Metrics

In causal Bayesian networks (CBNs),it is a fundamental problem to compute the causal effect of sum product.From the perspective of a directed acyclic graph,we show every CBN has a corresponding Bayesian network.Intervention is a fundamental operation in CBNs.Similar to Bayesian networks,CBNs also have the pruning strategy.After pruning the barren nodes,this paper devises an optimized jointree algorithm to compute the full atomic intervention on each node in a CBN.Then,this paper explores the multiple interventions on multiple nodes,and finds that multiple interventions have the commutative property.On the basis of the commutative property in multiple interventions,this paper proves the strategies,which can be used to optimize the computation of the causal effect of multiple interventions.Finally,we report experimental results to demonstrate the efficiency of our algorithm to compute the causal effects in CBNs.

Protein Solubility Prediction Based on Sequence Feature Fusion

NIU Fu-sheng, GUO Yan-bu, LI Wei-hua, LIU Wen-yang

Computer Science. 2022, 49 (1): 285-291. doi:10.11896/jsjkx.201100117

Abstract

PDF(2692KB) ( 978 )

References | Related Articles | Metrics

Protein solubility plays an important role in the research of drug design.Traditional biological experiments of detecting protein solubility are time-consuming and laborious.Identifying protein solubility based on computational methods has become an important research hot spot in bioinformatics.Aiming at the problem of insufficient representation of protein features by traditio-nal solubility prediction models,this paper designs a neural network model PSPNet based on protein sequence information and applies it to protein solubility prediction.PSPNet uses amino acid residue sequence embedding information and amino acid sequence evolution information to represent protein sequences.Then convolutional neural network is used to extract the local key information of amino acid sequence embedding features.Secondly,bidirectional LSTM network is used to extract the features of remote dependencies of protein sequences.Finally,the attention mechanism is used to fuse this feature and amino acid evolution information,and the fusion feature containing multiple sequence information is used in protein solubility prediction.The experimental results show that PASNet obtains the remarkable performance of protein solubility prediction compared with the benchmark me-thods and also has a good scalability.

Name Entity Recognition for Military Based on Domain Adaptive Embedding

LIU Kai, ZHANG Hong-jun, CHEN Fei-qiong

Computer Science. 2022, 49 (1): 292-297. doi:10.11896/jsjkx.201100007

Abstract

PDF(2065KB) ( 718 )

References | Related Articles | Metrics

In order to solve the poor quality problem of domain embedding space caused by inadequate military corpus which makes low accuracy of applying deep neural network model to military named entity recognition,this paper introduces a domain adaptive method to help learn the embedding of military fields from more useful information of additional fields through distributed representation of words.First,we establish the domain dictionary and combine CRF algorithm to perform domain adaptive word segment with the collected general domain and military areas corpus as training corpus for embedding,and word vectors are used as features and spliced with character vectors to enrich the embedding information and to validate the effect of word segmentation.Then the domain adaptive transformation is carried out to the heterogeneous embedded space of the general domain and the military domain,and the domain adaptive embedding is generated,as the input to BiLSTM-CRF layer of base model.At last,the recognition evaluation is carried out through CoNLL-2000.The experimental results show that,under the same model,the recognition precision rate (P),recall rate (R),and integrated F1 value (F1) of the proposed method are improved by 2.17%,1.04%,and 1.59%,respectively,compared with the military field embedding trained by a corpus which is obtained from general word segmentation.

Upper Confidence Bound Exploration with Fast Convergence

AO Tian-yu, LIU Quan

Computer Science. 2022, 49 (1): 298-305. doi:10.11896/jsjkx.201100194

Abstract

PDF(2606KB) ( 594 )

References | Related Articles | Metrics

Deep reinforcement learning method has achieved excellent results in large state space control tasks.Exploration has always been a research hotspot in this field.There are some problems in the existing exploration algorithms,such as blind exploration,and slow learning.To solve these problems,an upper confidence bound exploration with fast convergence (FAST-UCB) method is proposed.This method uses UCB method to explore the environment and improve the exploration efficiency.In order to alleviate the overestimation of Q value and balance the relationship between exploration and utilization,Q value clipped technique is added.Then,in order to balance the deviation and variance of the algorithm and make the agent learn quickly,the long short term memory unit is added to the network model,and an improved mixed monte carlo method is used to calculate the network error.Finally,FAST-UCB is applied to deep Q network,and compared with epsilon-greedy and UCB algorithms in control environment to verify its effectiveness.Besides,the proposed algorithm is compared with noise network exploration,bootstrapped exploration,asynchronous advantage actor critical algorithm and proximal policy optimization algorithm in Atari 2600 environment to verify its generalization.The experimental results show that FAST-UCB algorithm can achieve excellent results in these two environments.

Review on Video Privacy Protection

JIN Hua, ZHU Jing-yu, WANG Chang-da

Computer Science. 2022, 49 (1): 306-313. doi:10.11896/jsjkx.201200047

Abstract

PDF(1875KB) ( 938 )

References | Related Articles | Metrics

With the rapid development of video processing technology and the continuous reduction of hardware costs,surveillance equipment has become more and more widely used.The leakage of privacy issues brought about by the popularity of video surveillance has gradually become a research hotspot.According to the current research status in this field,video privacy protection methods are mainly divided into three stages:privacy subject identification,privacy subject protection and privacy information management.Each stage algorithm is classified and summarized,and its advantages and disadvantages are analyzed.The privacy region protection is an important part of this field.The protection methods are analyzed and compared in connection with the development of video coding.Finally,the existing problems in the field of video privacy protection are discussed,and the future research direction is prospected,which provides a reference for the related research of video privacy protection.

H.264/AVC Video Encryption Based on Adaptive Permutation of Macroblock Coding Information

LIANG Jian, HE Jun-hui

Computer Science. 2022, 49 (1): 314-320. doi:10.11896/jsjkx.201100089

Abstract

PDF(2396KB) ( 572 )

References | Related Articles | Metrics

The development of cloud storage makes people willing to upload personal video to the cloud,but the data security problems brought by it have become increasingly prominent,selective encryption is one of the effective ways to protect video privacy.Aiming at the problem of insufficient security in the current H.264/AVC video selective encryption method,a novel H.264/AVC video selective encryption method based on adaptive permutation of macroblock coding information is proposed.The method adaptively generates pseudo-random sequence frame by frame according to the macroblock types,uses the pseudo-random sequence to randomly permute the coded block pattern (CBP) and the residual data in the coding information of a macroblock between macroblocks,changes the intra prediction modes of I macroblocks,and flips the signs of motion vector differences of P macroblocks and B macroblocks.Experimental results show that the proposed method can preserve format compatibility with H.264/AVC coding standard,and has characteristics of large encryption space,good key sensitivity,and small video bitrate variation.Compared with the existing encryption schemes,the proposed method performs better in terms of visual security and resistance to state of the art of sketch attack.

Hierarchical Anonymous Voting Scheme Based on Threshold Ring Signature

FAN Jia-xing, WANG Zhi-wei

Computer Science. 2022, 49 (1): 321-327. doi:10.11896/jsjkx.201000032

Abstract

PDF(1577KB) ( 767 )

References | Related Articles | Metrics

Voting is a commonly used method in modern democratic society,involving many fields such as politics,stock companies,court decisions,etc.Voting is regarded as a specific form of balloting,with only two candidates in pro and con.Blockchain is a digital accounting technology with the characteristics of time stamp,openness and non-tamperability which satisfy the transpa-rency and verifiability of voting.In order to realize the anonymity of voting,this paper uses ring signature to hide the correspondence between voting content and the voter.This paper puts forward a hierarchical anonymous voting scheme,which realizes the legitimacy,confidentiality,non-repeatability,updateability and verifiability of voting.By creating a hierarchy mechanism for the voting of virtual identities,it can be used in situations where the votes vary from vote to each voter,and this agreement applies the threshold ring signature scheme to the voting scene for the first time,making the voting process simple and efficient for the final voting results once one party has more than half of the votes cast.

Visual Analysis Method of Blockchain Community Evolution Based on DPoS Consensus Mechanism

WEN Xiao-lin, LI Chang-lin, ZHANG Xin-yi, LIU Shang-song, ZHU Min

Computer Science. 2022, 49 (1): 328-335. doi:10.11896/jsjkx.201200118

Abstract

PDF(3537KB) ( 754 )

References | Related Articles | Metrics

DPoS (delegated proof of stake) is one of the current mainstream blockchain consensus mechanisms,and the unique node election mechanism makes it form an evolving blockchain community.Analyzing the evolution model of the blockchain community can discover the potential risks of the consensus mechanism,which has very important research significance.For the DPoS consensus mechanism blockchain data,a novel combination analysis method of the consensus mechanism effectiveness is proposed,and a set of visual analysis methods are designed to help users analyze the evolutionary model of the blockchain community from multiple angles.First,it quantifies the difference between the degree of completion of the work and the voting ranking before and after the node ranking change and analyzes the selection efficiency and incentive efficiency of the consensus mechanism;then,it focuses on the combined efficiency of the consensus mechanism,the evolution of the geographical distribution of nodes,and the comparison of the evolutionary differences between nodes and designs visual views and interactive means;finally,it designs and implements a visual analysis system of blockchain community evolution based on the DPoS consensus mechanism based on the real data of the EOS main chain and verifies the usability and effectiveness of this method through case studies and expert evaluation.

Novel Hash-time-lock-contract Based Cross-chain Token Swap Mechanism of Blockchain

LIU Feng, ZHANG Jia-hao, ZHOU Jun-jie, LI Mu, KONG De-li, YANG Jie, QI Jia-yin, ZHOU Ai-min

Computer Science. 2022, 49 (1): 336-344. doi:10.11896/jsjkx.210600170

Abstract

PDF(3482KB) ( 1513 )

References | Related Articles | Metrics

Blockchain is one of the technical hotspots in recent years,and the research theories related to it are being enriched.However,it is still facing the key problem of small throughput and low processing efficiency before it can be implemented into the industry.In response to this problem,cross-chain technology has been widely focused as a blockchain technology that has the potential to both expand throughput processing capacity and improve processing efficiency.This paper presents a novel hash-time-lock-contract based cross-chain token swap mechanism (NCASP),which creatively introduces an account system for the Fabric blockchain and integrates smart contract technology to achieve secure and seamless asset exchange between the Ethernet and Fabric blockchain network.The NCASP protocol sets up different intermediate accounts for asset escrow and transfer in each HTLC transfer,and destroys them in time after the transaction is completed,making the original cross-chain transaction rate unchanged while ensuring the security of the transaction.Simulation of the protocol shows that the protocol is applicable to the federated chain represented by Fabric and the public chains represented by Bitcoin and Ethernet,can achieve efficient and secure cross-chain asset exchange without the intervention of third-party blockchains,and can save about 26.8% in transaction efficiency compared with the cross-chain scheme of BSN (blockchain service network).The improved protocol extends the usage scenarios of the traditional HTLC cross-chain asset schemes,enabling the exchange of assets between different users with a balance of atomicity,fairness and transparency.

Blockchain Covert Communication Model for Plain Text Information Hiding

SHE Wei, HUO Li-juan, TIAN Zhao, LIU Wei, SONG Xuan

Computer Science. 2022, 49 (1): 345-352. doi:10.11896/jsjkx.201000112

Abstract

PDF(2089KB) ( 850 )

References | Related Articles | Metrics

Plain text information hiding is vulnerable to active attacks such as deletion and change,which makes the embedded secret information damaged.Blockchain is characterized by non-tampering,non-forgery,anonymity and node information synchronization,making it a natural platform for building hidden channels and ensuring that secret information is not destroyed.This paper proposes the blockchain covert communication model for plain text information hiding.Firstly,the location of the embedded secret information is determined according to the partial order relation.The sender uses the space method to embed the secret information into the plain text content.Then,a scenario of hidden communication in the blockchain network is constructed,and the sender publishes the transaction containing the plain text content to the blockchain network.Finally,after the transaction is packaged and a chain block is formed,any node can obtain the file as the receiver,but only the trusted party can extract the secret information through the inverse process of the embedded algorithm.Experimental comparison and analysis show that the model has better anti-detection,robustness,security and higher hiding capacity.More importantly,the blockchain as a channel enables the identity of the trusted party to be hidden,and the concealment of the communication process is doubly guaranteed.

Color Image Encryption Algorithm Based on Logistic-Sine-Cosine Mapping

ZHANG Sai-nan, LI Qian-mu

Computer Science. 2022, 49 (1): 353-358. doi:10.11896/jsjkx.201000041

Abstract

PDF(2541KB) ( 633 )

References | Related Articles | Metrics

The rapid development of technology has brought convenience for shooting and sharing images.However,with the ra-pid increase of image data,security problems such as leakage and tampering have frequently occurred.The application of image encryption technology is imminent.Especially the encryption of color images is in urgent need of improvement and development.The traditional encryption technology is mainly for data stream encryption,which is low in efficiency and large in calculation,ha-ving certain limitations.Based on the transformation domain encryption,the image is transformed from the spatial domain to the frequency domain for encryption,and then transformed to the spatial domain,which is a lossy encryption.Encryption based on chaos has a large key space,simple implementation,and fast encryption speed.However,multiple chaotic systems are generally required to enhance the security of encryption.For this reason,a simple and secure spatial encryption algorithm for the RGB three-channel color image is designed in this paper.The Logistic-Sine-Cosine mapping generates safer chaotic sequences.These chaotic sequences are used for four rounds of scrambling and spreading pixels.After a series of security analysis experiments,the security and effectiveness of the color image encryption algorithm based on Logistic-Sine-Cosine mapping have been verified.