Most Read articles

    Published in last 1 year |  In last 2 years |  In last 3 years |  All

    Published in last 1 year
    Please wait a minute...
    For Selected: Toggle Thumbnails
    Review of Deep Learning Applied to Time Series Prediction
    LIANG Hongtao, LIU Shuo, DU Junwei, HU Qiang, YU Xu
    Journal of Frontiers of Computer Science and Technology    2023, 17 (6): 1285-1300.   DOI: 10.3778/j.issn.1673-9418.2211108
    The time series is generally a set of random variables that are observed and collected at a certain frequency in the course of something??s development. The task of time series forecasting is to extract the core patterns from a large amount of data and to make accurate estimates of future data based on known factors. Due to the access of a large number of IoT data collection devices, the explosive growth of multidimensional data and the increasingly demanding requirements for prediction accuracy, it is difficult for classical parametric models and traditional machine learning algorithms to meet high efficiency and high accuracy requirements of prediction tasks. In recent years, deep learning algorithms represented by convolutional neural networks, recurrent neural networks and Trans-former models have achieved fruitful results in time series forecasting tasks. To further promote the development of time series prediction technology, common characteristics of time series data, evaluation indexes of datasets and models are reviewed, and the characteristics, advantages and limitations of each prediction algorithm are experimentally compared and analyzed with time and algorithm architecture as the main research line. Several time series prediction methods based on Transformer model are highlighted and compared. Finally, according to the problems and challenges of deep learning applied to time series prediction tasks, this paper provides an outlook on the future research trends in this direction.
    Reference | Related Articles | Metrics
    Research on Question Answering System on Joint of Knowledge Graph and Large Language Models
    ZHANG Heyi, WANG Xin, HAN Lifan, LI Zhao, CHEN Zirui, CHEN Zhe
    Journal of Frontiers of Computer Science and Technology    2023, 17 (10): 2377-2388.   DOI: 10.3778/j.issn.1673-9418.2308070
    The large language model (LLM), including ChatGPT, has shown outstanding performance in understanding and responding to human instructions, and has a profound impact on natural language question answering (Q&A). However, due to the lack of training in the vertical field, the performance of LLM in the vertical field is not ideal. In addition, due to its high hardware requirements, training and deploying LLM remains difficult. In order to address these challenges, this paper takes the application of traditional Chinese medicine formulas as an example, collects the domain related data and preprocesses the data. Based on LLM and knowledge graph, a vertical domain Q&A system is designed. The system has the following capabilities: (1) Information filtering. Filter out vertical domain related questions and input them into LLM to answer. (2) Professional Q&A. Generate answers with more professional knowledge based on LLM and self-built knowledge base. Compared with the fine-tuning method of introducing professional data, using this technology can deploy large vertical domain models without the need for retraining. (3) Extract conversion. By strengthening the information extraction ability of LLM and utilizing its generated natural language responses, structured knowledge is extracted and matched with a professional knowledge graph for professional verification. At the same time, structured knowledge can be transformed into readable natural language, achieving a deep integration of large models and knowledge graphs. Finally, the effect of the system is demonstrated and the performance of the system is verified from both subjective and objective perspectives through two experiments of subjective evaluation of experts and objective evaluation of multiple choice questions.
    Reference | Related Articles | Metrics
    Review of Medical Image Segmentation Based on UNet
    XU Guangxian, FENG Chun, MA Fei
    Journal of Frontiers of Computer Science and Technology    2023, 17 (8): 1776-1792.   DOI: 10.3778/j.issn.1673-9418.2301044
    As one of the most important semantic segmentation frameworks in convolutional neural networks (CNN), UNet is widely used in image processing tasks such as classification, segmentation, and target detection of medical images. In this paper, the structural principles of UNet are described, and a comprehensive review of UNet-based networks and variant models is presented. The model algorithms are fully investigated from several perspectives, and an attempt is made to establish an evolutionary pattern among the models. Firstly, the UNet variant models are categorized according to the seven medical imaging systems they are applied to, and the algorithms with similar core composition are compared and described. Secondly, the principles, strengths and weaknesses, and applicable scenarios of each model are analyzed. Thirdly, the main UNet variant networks are summarized in terms of structural principles, core composition, datasets, and evaluation metrics. Finally, the inherent shortcomings and solutions of the UNet network structure are objectively described in light of the latest advances in deep learning, providing directions for continued improvement in the future. At the same time, other technological evolutions and application scenarios that can be combined with UNet are detailed, and the future development trend of UNet-based variant networks is further envisaged.
    Reference | Related Articles | Metrics
    Survey of Research on Instance Segmentation Methods
    HUANG TAO, LI Hua, ZHOU Gui, LI Shaobo, WANG Yang
    Journal of Frontiers of Computer Science and Technology    2023, 17 (4): 810-825.   DOI: 10.3778/j.issn.1673-9418.2209051
    In recent years, with the continuous improvement of computing level, the research of instance segment-ation methods based on deep learning has made great breakthroughs. Image instance segmentation can distinguish different instances of the same class in images, which is an important research direction in the field of computer vision with broad research prospects, and has achieved great actual application value in scene comprehension, medical image analysis, machine vision, augmented reality, image compression, video monitoring, etc. Recently, instance segmentation methods have been updated more and more frequently, but there is a little literature to comprehensively and systematically analyze the research background related to instance segmentation. This paper  provides a comprehensive and systematic analysis and summary of the image instance segmentation methods based on deep learning. Firstly, this paper introduces the currently used common public datasets and evaluation indexes in instance segmentation, and analyzes the challenges of current datasets. Secondly, this paper respectively combs and summarizes the instance segmentation algorithms in the characteristics of two-stage segmentation methods and single-stage segmentation methods, elaborates their central ideas and design thoughts, and summarizes the advantages and shortcomings of the two types of methods. Thirdly, this paper evaluates the segmentation accuracy and speed of the models on a public dataset. Finally, this paper summarizes the current difficulties and challenges of instance segmentation, presents the solution ideas for facing the challenges, and makes a prospect for future research directions.
    Reference | Related Articles | Metrics
    Survey of Knowledge Graph Recommendation System Research
    ZHAO Yehui, LIU Lin, WANG Hailong, HAN Haiyan, PEI Dongmei
    Journal of Frontiers of Computer Science and Technology    2023, 17 (4): 771-791.   DOI: 10.3778/j.issn.1673-9418.2205052
    Recommendation systems can obtain user preferences in massive data information to better achieve personalized recommendations, improve user physical examination and solve information overload on the Internet. However, it still suffers from cold start and data sparsity problems. A knowledge graph, as a structured knowledge base with a large number of entities and rich semantic relationships, can not only improve the accuracy of recommendation systems, but also provide the interpretability for recommendation items, thus enhancing users?? trust in recommendation systems, and providing new methods and ideas to solve a series of key problems in recommendation systems. This paper firstly studies and analyzes knowledge graph recommendation systems, classifies them into multi-domain knowledge graph recommendation systems and domain-specific knowledge graph recommendation systems based on the classification of application fields, and further classifies them according to the characteristics of these knowledge graph recommendation methods, and conducts quantitative and qualitative analyses for each type of methods. Secondly, this paper lists the datasets commonly used by knowledge graph recommendation systems in the application fields, and gives an overview of the size and characteristics of the datasets. Finally, this paper outlooks and summarizes the future research directions of knowledge graph recommendation systems.
    Reference | Related Articles | Metrics
    Survey on Backdoor Attacks and Countermeasures in Deep Neural Network
    QIAN Hanwei, SUN Weisong
    Journal of Frontiers of Computer Science and Technology    2023, 17 (5): 1038-1048.   DOI: 10.3778/j.issn.1673-9418.2210061
    The neural network backdoor attack aims to implant a hidden backdoor into the deep neural network, so that the infected model behaves normally on benign test samples, but behaves abnormally on poisoned test samples with backdoor triggers. For example, all poisoned test samples will be predicted as the target label by the infected model. This paper provides a comprehensive review and the taxonomy for existing attack methods according to the attack objects, which can be categorized into four types, including data poisoning attacks, physical world attacks, model poisoning attacks, and others. This paper summarizes the existing backdoor defense technologies from the perspective of attack and defense confrontation, which include poisoned sample identifying, poisoned model identifying, poisoned test sample filtering, and others. This paper explains the principles of deep neural network backdoor defects from the perspectives of deep learning mathematical principles and visualization, and discusses the difficulties and future development directions of deep neural network backdoor attacks and countermeasures from the perspectives of software engineering and program analysis. It is hoped that this survey can help researchers understand the research progress of deep neural network backdoor attacks and countermeasures, and provide more inspiration for designing more robust deep neural networks.
    Reference | Related Articles | Metrics
    Research Review of Image Semantic Segmentation Method in High-Resolution Remote Sensing Image Interpretation
    MA Yan, Gulimila·Kezierbieke
    Journal of Frontiers of Computer Science and Technology    2023, 17 (7): 1526-1548.   DOI: 10.3778/j.issn.1673-9418.2211015
    Rapid acquisition of remote sensing information has important research significance for the development of image semantic segmentation methods in remote sensing image interpretation applications. With more and more types of data recorded by satellite remote sensing images and more and more complex feature information, accurate and effective extraction of information in remote sensing images has become the key to interpret remote sensing images by image semantic segmentation methods. In order to explore the image semantic segmentation method for fast and efficient interpretation of remote sensing images, a large number of image semantic segmentation methods for remote sensing images are summarized. Firstly, the traditional image semantic segmentation methods are reviewed and divided into edge detection-based segmentation methods, region-based segmentation methods, threshold-based segmentation methods and segmentation methods combined with specific theories. At the same time, the limitations of traditional image semantic segmentation methods are analyzed. Secondly, the semantic segmentation methods based on deep learning are elaborated in detail, and the basic ideas and technical characteristics of each method are used as the classification criteria. They are divided into four categories: FCN-based methods, codec-based methods, dilated convolution-based methods and attention-based methods. The sub-methods contained in each type of method are summarized, and the advantages and disadvantages of these methods are compared and analyzed. Then, the common datasets and performance evaluation indexes of remote sensing image semantic segmentation are briefly introduced. Experimental results of classical network models on different datasets are given, and the performance of different models is evaluated. Finally, the challenges of image semantic segmentation methods in high-resolution remote sensing image interpretation are analyzed, and the future development trend is prospected.
    Reference | Related Articles | Metrics
    Survey of Causal Inference for Knowledge Graphs and Large Language Models
    LI Yuan, MA Xinyu, YANG Guoli, ZHAO Huiqun, SONG Wei
    Journal of Frontiers of Computer Science and Technology    2023, 17 (10): 2358-2376.   DOI: 10.3778/j.issn.1673-9418.2307065
    In recent decades, causal inference has been a significant research topic in various fields, including statistics, computer science, education, public policy, and economics. Most causal inference methods focus on the analysis of sample observational data and text corpora. However, with the emergence of various knowledge graphs and large language models, causal inference tailored to knowledge graphs and large models has gradually become a research hotspot. In this paper, different causal inference methods are classified based on their orientation towards sample observational data, text data, knowledge graphs, and large language models. Within each classification, this paper provides a detailed analysis of classical research works, including their problem definitions, solution methods, contributions, and limitations. Additionally, this paper places particular emphasis on discussing recent advancements in the integration of causal inference methods with knowledge graphs and large language models. Various causal inference methods are analyzed and compared from the perspectives of efficiency and cost, and specific applications of knowledge graphs and large language models in causal inference tasks are summarized. Finally, future development directions of causal inference in combination with knowledge graphs and large models are prospected.
    Reference | Related Articles | Metrics
    Review on Multi-lable Classification
    LI Dongmei, YANG Yu, MENG Xianghao, ZHANG Xiaoping, SONG Chao, ZHAO Yufeng
    Journal of Frontiers of Computer Science and Technology    2023, 17 (11): 2529-2542.   DOI: 10.3778/j.issn.1673-9418.2303082
    Multi-label classification refers to the classification problem where multiple labels may coexist in a single sample. It has been widely applied in fields such as text classification, image classification, music and video classification. Unlike traditional single-label classification problems, multi-label classification problems become more complex due to the possible correlation or dependence among labels. In recent years, with the rapid development of deep learning technology, many multi-label classification methods combined with deep learning have gradually become a research hotspot. Therefore, this paper summarizes the multi-label classification methods from the traditional and deep learning-based perspectives, and analyzes the key ideas, representative models, and advantages and disadvantages of each method. In traditional multi-label classification methods, problem transformation methods and algorithm adaptation methods are introduced. In deep learning-based multi-label classification methods, the latest multi-label classification methods based on Transformer are reviewed particularly, which have become one of the mainstream methods to solve multi-label classification problems. Additionally, various multi-label classification datasets from different domains are introduced, and 15 evaluation metrics for multi-label classification are briefly analyzed. Finally, future work is discussed from the perspectives of multi-modal data multi-label classification, prompt learning-based multi-label classification, and imbalanced data multi-label classification, in order to further promote the development and application of multi-label classification.
    Reference | Related Articles | Metrics
    Survey of Fake News Detection with Multi-model Learning
    LIU Hualing, CHEN Shanghui, CAO Shijie, ZHU Jianliang, REN Qingqing
    Journal of Frontiers of Computer Science and Technology    2023, 17 (9): 2015-2029.   DOI: 10.3778/j.issn.1673-9418.2301064
    While social media brings convenience to people, it has also become a channel for the arbitrary spread of fake news. If not detected and stopped in time, it is easy to cause public panic and social unrest. Therefore, exploring accurate and efficient fake news detection technology has high theoretical value and practical significance. This paper provides a comprehensive overview of the related fake news detection techniques. Firstly, the relevant concepts of multi-modal fake news are sorted and summarized, and the trend of changes in single-modal and multi-modal news datasets is analyzed. Secondly, this paper introduces single-modal fake news detection techniques based on machine learning and deep learning, which have been widely used in the field of fake news detection. However, traditional single-modal techniques cannot fully explore the deep logic of fake news because fake news usually contains multiple data presentations. Thus, they are unable to effectively deal with the challenges brought by multi-modal fake news data. To address this issue, this paper summarizes and discusses advanced multi-modal fake news detection techniques from the perspectives of multi-stream and graph architectures, and explores their concepts and potential drawbacks. Finally, this paper analyzes the difficulties and bottlenecks in the current research on fake news detection and provides future research directions based on these analyses.
    Reference | Related Articles | Metrics
    Research Progresses of Multi-modal Intelligent Robotic Manipulation
    ZHANG Qiuju, LYU Qing
    Journal of Frontiers of Computer Science and Technology    2023, 17 (4): 792-809.   DOI: 10.3778/j.issn.1673-9418.2212070
    The flexible production trend in manufacturing and the diversified expansion of applications in service industry have prompted fundamental changes in the application demands of robots. The uncertainty of tasks and environments imposes higher requirements on the intelligence of robotic manipulation. The use of multi-modal information to monitor robotic manipulation can effectively improve the intelligence and flexibility of robot. This paper provides an in-depth analysis of the role of multi-modal information in enhancing the intelligence of robotic manipulation from the perspective of multi-modal information fusion on the basis of the two key issues of manipulation cognition and manipulation control. Firstly, the concepts of intelligent robotic manipulation and multi-modal information are clarified, and the merits of applying multi-modal information are also introduced. Then, the commonly used perception models and control methods are deeply analyzed and the existing work is sorted out and introduced in a systematic way. According to different levels of perception goals, robotic manipulation perception is divided into object perception, constraint perception and state perception; according to different control methods, the most commonly used control fusion based on analysis model, imitation learning control and reinforcement learning control are introduced. Finally, the current technical challenges and potential development trends are also discussed.
    Reference | Related Articles | Metrics
    Overview of Blockchain and Database Fusion
    LI Xinhang, LI Chao, ZHANG Guigang, XING Chunxiao
    Journal of Frontiers of Computer Science and Technology    2023, 17 (4): 761-770.   DOI: 10.3778/j.issn.1673-9418.2211021
    Recently, blockchain has been increasingly applied in finance and other fields, which is designed with the purpose to achieve secure and trusted data storage while reducing the efficiency. Meanwhile, its insufficient technological accumulation causes technology support shortage, which limits its development. As another data storage technology with development history for decades, database has a series of comparatively mature and complete technologies. As two widely-used data storage technologies, it is a significative research direction to integrate these two technologies through system architecture design and achieve a new generation of data storage technology. Based on the description of the design concept, technical characteristics and overall architecture of blockchain and database technology, the commonalities and differences as well as the advantages and disadvantages between these two data storage techniques are analyzed from several perspectives. From these perspectives and taking the two technology integration paradigms of out-of-the blockchain database and out-of-the database blockchain as the benchmark, this paper summarizes the existing works of blockchain and database technology integration. Furthermore, the crucial problems and  future development directions of blockchains and database technology integration are analyzed from a variety of dimensions.
    Reference | Related Articles | Metrics
    YOLOv8-VSC: Lightweight Algorithm for Strip Surface Defect Detection
    WANG Chunmei, LIU Huan
    Journal of Frontiers of Computer Science and Technology    2024, 18 (1): 151-160.   DOI: 10.3778/j.issn.1673-9418.2308060
    Currently, in the field of strip steel surface defect detection, the generalized target detection algorithm is highly complex and computationally large, while terminal equipment responsible for the detection of some small and medium-sized enterprises usually does not have strong computational capabilities, and the computational resources are limited, which leads to difficulties in the deployment of detection algorithms. To solve this problem, this paper proposes a lightweight strip steel surface defect detection model YOLOv8-VSC based on the YOLOv8n target detec-tion framework, which uses a lightweight VanillaNet network as the backbone feature extraction network and reduces the complexity of the model by reducing the unnecessary branching structure. Meanwhile, the SPD module is introduced to speed up the inference of the model while reducing the number of network layers. To further improve the detection accuracy, a lightweight up-sampling operator, CARAFE, is used in the feature fusion network to improve the quality and richness of the features. Finally, extensive experiments on the NEU-DET dataset yield a model with parametric and computational quantities of 1.96×106 and 6.0 GFLOPs, which are only 65.1% and 74.1% of the baseline, and the mAP reaches 80.8%, which is an improvement of 1.8 percentage points from the baseline. In addition, experimental results on the aluminum surface defect dataset and the VOC2012 dataset show that the proposed algorithm has good robustness. Compared with advanced target detection algorithms, the proposed algorithm requires fewer computational resources while ensuring high detection accuracy.
    Reference | Related Articles | Metrics
    Advances in Knowledge Graph Embedding Based on Graph Neural Networks
    YAN Zhaoyao, DING Cangfeng, MA Lerong, CAO Lu, YOU Hao
    Journal of Frontiers of Computer Science and Technology    2023, 17 (8): 1793-1813.   DOI: 10.3778/j.issn.1673-9418.2212063
    As graph neural networks continue to develop, knowledge graph embedding methods based on graph neural networks are receiving increasing attention from researchers. Compared with traditional methods, they can better handle the diversity and complexity of entities, and capture the multiple features and complex relationships of entities, thereby improving the representation ability and application value of knowledge graphs. This paper firstly outlines the development history of knowledge graphs and the basic concepts of knowledge graphs and graph neural networks. Secondly, it focuses on discussing the design ideas and algorithm frameworks of knowledge graph embedding based on graph convolution, graph neural networks, graph attention, and graph autoencoders. Then, it describes the performance of graph neural network knowledge graph embedding in tasks such as link prediction, entity alignment, knowledge graph reasoning, and knowledge graph completion, while supplementing some research on commonsense knowledge graphs with graph neural networks. Finally, this paper makes a comprehensive summary, and future research directions are outlined with respect to some challenges and issues in knowledge graph embedding.
    Reference | Related Articles | Metrics
    Research Progress and Prospect of Ring Signatures
    XIE Jia, LIU Shizhao, WANG Lu, GAO Juntao, WANG Baocang
    Journal of Frontiers of Computer Science and Technology    2023, 17 (5): 985-1001.   DOI: 10.3778/j.issn.1673-9418.2210022
    As a special group signature, ring signature has been widely used in anonymous voting, anonymous deposit and anonymous transaction because it can not only complete the signature without the cooperation of ring members, but also ensure the anonymity of the signer. Firstly, this paper takes time as the main line, divides the development of ring signatures into different stages, and divides ring signatures into threshold ring signatures, linkable ring signatures, ring signatures with revocable anonymity, and repudiable ring signatures according to the attributes in each stage. Through the analysis of the development process of ring signature, it can be seen that the research progress of ring signature in the field of threshold ring signature and linkable ring signature is prominent, and their application fields are also the most extensive. In the post-quantum era, cryptographic schemes based on traditional number theory problems such as large integer factorization and discrete logarithms are no longer secure, and lattice-based public key cryptography has become the best candidate for cryptographic standards in the post-quantum era because of its advantages such as quantum-immune, the reduction of the worst-case to average-case, and so on. Therefore, this paper focuses on the detailed analysis and efficiency comparison of existing lattice-based threshold ring signatures and lattice-based linkable ring signatures. The inherent anonymity of ring signatures makes them have unique advantages in the era of industrial blockchain, so this paper elaborates on several applications of ring signatures in blockchain. For example, the application of ring signature in anonymous voting, medical data sharing, and Internet of vehicles is summarized and analyzed. The application significance of ring signature in the fields of virtual currency, SIP cloud call protocol, and Ad Hoc network is briefly sorted out. Finally,  the research of ring signature technology in recent years is analyzed, and the current problems are summarized.
    Reference | Related Articles | Metrics
    Survey on Personality Recognition Based on Social Media Data
    LIN Hao, WANG Chundong, SUN Yongjie
    Journal of Frontiers of Computer Science and Technology    2023, 17 (5): 1002-1016.   DOI: 10.3778/j.issn.1673-9418.2212012
    Personality is a stable construction, which is associated with thoughts, emotions and behaviors of human. Any technology involved in understanding, analyzing and predicting human behaviors may benefit from personality recognition. Accurately recognizing the personality will contribute to the research of human-computer interaction, recommendation system, cyberspace security, etc. Social media provides high-quality data for personality recog-nition, while classic personality measurement methods such as self-report scales and projective tests can no longer match the social media data in the age of big data. Moreover, the current mainstream personality recognition methods based on machine learning still have lots of room for performance improvement. Therefore, this paper investigates the literature on personality recognition based on social media data, introduces the background knowledge of personality recognition and summarizes the research status according to the data types of recognition model input used, specifically based on social text data, social image data, social application data and multimodal data. Finally, this paper proposes seven future research directions of personality recognition based on social media data.
    Reference | Related Articles | Metrics
    Named Entity Recognition Method of Large Language Model for Medical Question Answering System
    YANG Bo, SUN Xiaohu, DANG Jiayi, ZHAO Haiyan, JIN Zhi
    Journal of Frontiers of Computer Science and Technology    2023, 17 (10): 2389-2402.   DOI: 10.3778/j.issn.1673-9418.2307061
    In medical question answering systems, entity recognition plays a major role. Entity recognition based on deep learning has received more and more attention. However, in the medical question answering system, due to the lack of annotated training data, deep learning methods cannot well identify discontinuous and nested entities in medical text. Therefore, a large language model-based entity recognition application method is proposed, and it is applied to the medical problem system. Firstly, the dataset related to medical question answering is processed into text that can be analyzed and processed by a large language model. Secondly, the output of the large language model is classified, and different classifications are processed accordingly. Then, the input text is used for intent recognition, and  finally the results of entity recognition and intent recognition are sent to the medical knowledge graph for query, and the answer to the medical question and answer is obtained. Experiments are performed on 3 typical datasets and compared with several typical correlation methods. The results show that the method proposed in this paper performs better.
    Reference | Related Articles | Metrics
    Review of Research on Fatigue Driving Detection Based on Driver Facial Features
    YANG Yanyan, LI Leixiao, LIN Hao
    Journal of Frontiers of Computer Science and Technology    2023, 17 (6): 1249-1267.   DOI: 10.3778/j.issn.1673-9418.2208041
    Fatigue driving is one of the main factors that threaten the safety of drivers and traffic. Efficient and accurate fatigue driving detection method can effectively ensure the safety of drivers and their surrounding traffic, maintain traffic order, and reduce property losses and casualties. The fatigue driving detection method based on the driver’s physiological characteristics and vehicle driving information has many limitations, such as being unfriendly to the driver and having numerous influencing factors. Therefore, the fatigue driving detection method based on the driver’s facial features has become a research hotspot. Firstly, the facial feature performance of fatigue driving is described, and advantages, disadvantages and application scenarios of common public datasets in the field of fatigue driving are summarized. Secondly, the advantages and disadvantages of common face detection algorithms in the field of fatigue driving detection are analyzed and studied by using open datasets and comparative experiments. Then, this paper generalizes the process of detection methods based on driver facial features, whose methods and technologies used in the key steps are reviewed. Furthermore, fatigue discriminant parameters and methods of fatigue driving results prediction are summarized. Finally, this paper ends with a discussion of current challenges of fatigue driving detection methods based on driver facial features and looks forward to the future research.
    Reference | Related Articles | Metrics
    Research on Corn Disease Detection Based on Improved YOLOv5 Algorithm
    SU Junkai, DUAN Xianhua, YE Zhaobing
    Journal of Frontiers of Computer Science and Technology    2023, 17 (4): 933-941.   DOI: 10.3778/j.issn.1673-9418.2210066
    In order to solve the problems of backward identification technology, low efficiency and insufficient accuracy of corn leaf disease, an improved YOLOv5 algorithm is proposed to identify corn disease. In order to maintain low computation of the model, and improve detection speed and algorithm performance, the feature extraction structure of traditional YOLOv5s network is improved, and CA (coordinate attention) attention mechanism is added to the backbone network, which improves the problem of target undetected, and helps the model locate and identify more accurately. In the neck, BiFPN (bidirectional feature pyramid network) is used to replace original PANet (path aggregation network), and the application of multi-scale semantic features is improved through two-way feature fusion to enhance the extraction for deep features of images. A small target monitoring layer is added to enhance the detection effect of small objects. Loss function is improved and Focal-EIOU Loss is introduced to improve the precision of BBox regression. Compared with traditional YOLOv5s, Recall is increased by 4.61 percentage points, AP is increased by 4.5 percentage points, mAP@0.5 is increased by 2.14 percentage points, and detection speed is increased by 4.5 FPS. Experimental results show that the improved YOLOv5 algorithm significantly improves the efficiency and performance of the algorithm with little complexity increase, and the effect is better than traditional YOLOv5s algorithm.
    Reference | Related Articles | Metrics
    Word Embedding Methods in Natural Language Processing: a Review
    ZENG Jun, WANG Ziwei, YU Yang, WEN Junhao, GAO Min
    Journal of Frontiers of Computer Science and Technology    2024, 18 (1): 24-43.   DOI: 10.3778/j.issn.1673-9418.2303056
    Word embedding, as the first step in natural language processing (NLP) tasks, aims to transform input natural language text into numerical vectors, known as word vectors or distributed representations, which artificial intelligence models can process. Word vectors, the foundation of NLP tasks, are a prerequisite for accomplishing various NLP downstream tasks. However, most existing review literature on word embedding methods focuses on the technical routes of different word embedding methods, neglecting comprehensive analysis of the tokenization methods and the complete evolutionary trends of word embedding. This paper takes the introduction of the word2vec model and the Transformer model as pivotal points. From the perspective of whether generated word vectors can dynamically change their implicit semantic information to adapt to the overall semantics of input sentences, this paper categorizes word embedding methods into static and dynamic approaches and extensively discusses this classification. Simultaneously, it compares and analyzes tokenization methods in word embedding, including whole and sub-word segmentation. This paper also provides a detailed enumeration of the evolution of language models used to train word vectors, progressing from probability language models to neural probability language models and the current deep contextual language models. Additionally, this paper summarizes and explores the training strategies employed in pre-training language models. Finally, this paper concludes with a summary of methods for evaluating word vector quality, an analysis of the current state of word embedding methods, and a prospective outlook on their development.
    Reference | Related Articles | Metrics
    Review of Privacy-Preserving Research in Recommendation Systems
    FENG Han, YI Huawei, LI Xiaohui, LI Rui
    Journal of Frontiers of Computer Science and Technology    2023, 17 (8): 1814-1832.   DOI: 10.3778/j.issn.1673-9418.2211069
    The recommendation system needs to extract the historical data information of the relevant users on a large scale as the training set of the prediction model. The larger and more specific the amount of data provided by users is, the easier it is for the personal information to be inferred, which is easy to lead to the leakage of personal privacy, so that the user’s trust in the service provider is reduced and the relevant data are no longer provided for the system, resulting in the reduction of the system recommendation accuracy or even more difficult to complete the recommendation. Therefore, how to obtain user information for effective recommendation with high accuracy under the premise of protecting user privacy has become a research hotspot. This paper firstly summarizes the privacy-preserving technology, including differential privacy technology, homomorphic encryption technology, federated learning and secure multi-party computing technology, and compares these commonly used privacy-preserving tech-nologies. Then, from the perspective of balancing the relationship between privacy injection and recommendation accuracy, the privacy-preserving technologies adopted by the user side, the server side and the user-server side are introduced, and the research results of privacy-preserving in recommendation systems at home and abroad are systematically elaborated. Based on this, a summary, comparison, and analysis are conducted. Next, the experi-mental results of the recommendation algorithm based on differential privacy technology are compared, and the shortcomings of the corresponding technology are analyzed. Finally, the future development directions of the recommendation system based on privacy-preserving are prospected.
    Reference | Related Articles | Metrics
    Survey on Blockchain-Based Cross-Domain Authentication for Internet of Things Terminals
    HUO Wei, ZHANG Qionglu, OU Wei, HAN Wenbao
    Journal of Frontiers of Computer Science and Technology    2023, 17 (9): 1995-2014.   DOI: 10.3778/j.issn.1673-9418.2211004
    Internet of things (IoT) devices are widely distributed, numerous and complex, which are involved in multiple management domains. They are often in uncontrollable environments and are more vulnerable to attacks than traditional Internet terminals, the security management and protection of IoT terminals face greater risks and challenges. As “the first line of defense” for the security protection of IoT devices, authentication plays an irreplaceable and important role in the development of IoT security. The blockchain technology has the characteristics and advantages of decentralization, distribution, immutability and traceability. And thus, it can effectively solve the single-point trust failure of trusted third parties and satisfy the principle of least authorization for multi-domain heterogeneity in cross-domain authentication for IoT terminals. Using the blockchain technology is an important trend in the future development of the IoT device cross-domain authentication. This paper categorizes and summarizes the main research achievements of IoT cross-domain authentication based on blockchain technology in recent years according to three categories: integrating traditional identity authentication mechanisms such as PKI and IBS/IBC, adopting inter-blockchain technology, and other cross-domain authentication technologies based on blockchain. Then this paper analyzes the technical characteristics, advantages and disadvantages of each different scheme. On this basis, the current problems and issues in the field of cross-domain authentication of IoT devices are summarized, and the future research directions and development suggestions for the cross-domain authentication of IoT terminals are given, so as to achieve a general and overall grasp of the research progress and development trend of IoT device cross-domain authentication schemes based on blockchain technology.
    Reference | Related Articles | Metrics
    Survey on Few-Shot Knowledge Graph Completion Technology
    PENG Yanfei, ZHANG Ruisi, WANG Ruihua, GUO Jialong
    Journal of Frontiers of Computer Science and Technology    2023, 17 (6): 1268-1284.   DOI: 10.3778/j.issn.1673-9418.2209069
    Few-shot knowledge graph completion (FKGC) is a new research hotspot in the field of knowledge graph completion, which aims to complete knowledge graph with a few samples of data. This task is of great importance in practical application and the fields of knowledge graph. In order to further promote the development of the field of FKGC, this paper summarizes and analyzes the current methods. Firstly, this paper describes the concept of FKGC and related content. Secondly, three types of FKGC methods are summarized based on technical methods, including scale learning-based methods, meta learning-based methods, and other model-based methods. In addition, this paper analyzes and summarizes each method from the perspectives of model core, model ideas, advantages and disadvantages, etc. Then, the datasets and evaluation indexes of FKGC method are summarized, and the FKGC method is analyzed from two aspects of model characteristics and experimental results. Finally, starting from the practical problems, this paper summarizes the difficult problems of the current FKGC task, analyses the difficulties behind the problems, gives the corresponding solutions, and prospects several development directions that deserve attention in this field in the future.
    Reference | Related Articles | Metrics
    Survey on Inductive Learning for Knowledge Graph Completion
    LIANG Xinyu, SI Guannan, LI Jianxin, TIAN Pengxin, AN Zhaoliang, ZHOU Fengyu
    Journal of Frontiers of Computer Science and Technology    2023, 17 (11): 2580-2604.   DOI: 10.3778/j.issn.1673-9418.2303063
    Knowledge graph completion can make knowledge graph more complete. However, traditional knowledge graph completion methods assume that all test entities and relations appear in the training process. Due to the evolving nature of real world KG, once unseen entities or relations appear, the knowledge graph needs to be retrained. Inductive learning for knowledge graph completion aims to complete triples containing unseen entities or unseen relations without training the knowledge graph from scratch, so it has received much attention in recent years. Firstly, starting from the basic concept of knowledge graph, this paper divides knowledge graph completion into two categories: transductive and inductive. Secondly, from the theoretical perspective of inductive knowledge graph completion, it is divided into two categories: semi-inductive and fully-inductive, and the models are summarized from this perspective. Then, from the technical perspective of inductive knowledge graph completion, it is divided into two categories: based on structural information and based on additional information. The methods based on structural information are subdivided into three categories: based on inductive embedding, based on logical rules and based on meta learning, and the methods based on additional information are subdivided into two categories: based on text information and other information. The current methods are further subdivided, analyzed and compared. Finally, it forecasts the main research directions in the future.
    Reference | Related Articles | Metrics
    Review of Visual Question Answering Technology
    WANG Yu, SUN Haichun
    Journal of Frontiers of Computer Science and Technology    2023, 17 (7): 1487-1505.   DOI: 10.3778/j.issn.1673-9418.2303025
    Visual question answering (VQA) is a popular cross-modal task that combines natural language pro-cessing and computer vision techniques. The main objective of this task is to enable computers to intelligently recognize and retrieve visual content and provide accurate answers. VQA involves the integration of multiple technologies such as object recognition and detection, intelligent question answering, image attribute classification, and scene analysis. It can support a wide range of cutting-edge interactive AI tasks such as visual dialogue and visual navigation, and has broad application prospects and great value. Over the past few years, the development of computer vision, natural language processing, and cross-modal AI models has provided many new technologies and methods for achieving the task of visual question answering. This paper mainly summarizes the mainstream models and specialized datasets in the field of visual question answering between 2019 and 2022. Firstly, this paper provides a review and discussion of the mainstream technical methods used in the key steps of implementing the visual question answering task, based on the module framework. Next, it subdivides various types of models in this field according to the technical methods adopted by mainstream models, and briefly introduces their improvement focus and limitations. Then, it summarizes the commonly used datasets and evaluation metrics for visual question answering, and compares and discusses the performance of several typical models. Finally, this paper focuses on the key issues that need to be addressed in the current visual question answering field, and predicts and prospects the future application and technological development in this field.
    Reference | Related Articles | Metrics
    Traffic Prediction Method Integrating Graph Wavelet and Attention Mechanism
    XUE Yanming, LI Guanghui, QI Tao
    Journal of Frontiers of Computer Science and Technology    2023, 17 (6): 1405-1416.   DOI: 10.3778/j.issn.1673-9418.2111012
    Traffic predicting is a critical component of modern intelligent transportation systems for traffic management and control. However, the traffic flow is complex. On one hand, the urban road structure is highly correlative, and there often exists a nonlinear structural dependence between different roads. On the other hand, traffic flow data often change dynamically over time. In recent years, many studies have tried to use deep learning methods to extract complex structural features in traffic flow. However, the process of local feature extraction still lacks flexibility, and ignores the dynamic variability as well as the correlation of spatio-temporal features. To this end, this paper proposes a new traffic prediction method integrating graph wavelet and attention mechanism. This method uses wavelet transform and an adaptive matrix to extract local and global spatial features of traffic flow respectively, and combines the improved recurrent neural network to extract local temporal characteristic information. Meanwhile, the attention mechanism is adopted in this method to capture the temporal and spatial dynamic variability. Then this method applies a spatio-temporal feature fusion mechanism to fusing local and global temporal and spatial features. Experimental results show that this method can extract spatial and temporal features of real traffic datasets well, and it outperforms the existing methods.
    Reference | Related Articles | Metrics
    Lightweight and High-Precision Dual-Channel Attention Mechanism Module
    CHEN Xiaolei, LU Yubing, CAO Baoning, LIN Dongmei
    Journal of Frontiers of Computer Science and Technology    2023, 17 (4): 857-867.   DOI: 10.3778/j.issn.1673-9418.2112052
    At present, most attention mechanism modules improve the application accuracy of deep learning models, but also bring the defect of increased model complexity. In response to this problem, this paper proposes a lightweight and efficient dual-channel attention mechanism module (EDCA). EDCA compresses and rearranges the feature maps in three directions of channel, width, and height. One-dimensional convolution is used to obtain the combined weight information, and then the weight information is segmented and applied to corresponding dimensions to obtain feature attention. This paper conducts a full experiment on EDCA on image classification dataset miniImageNet and target detection dataset Pascal VOC2007. Compared with SENet, CBAM, SGE, ECA-Net and Coordinate Attention, EDCA requires less computation and parameters. When ResNet50+EDCA is used on miniImageNet dataset, the Top-1 accuracy is improved by 0.0243, 0.0218, 0.0221, 0.0225 and 0.0141, respectively. When MobileNetV3+YOLOv4+EDCA is used on Pascal VOC2007 dataset, AP50 is improved by 0.0094, 0.0046, 0.0059 and 0.0014 compared with SENet, CBAM, ECA-Net and Coordinate Attention, respectively.
    Reference | Related Articles | Metrics
    Review of Research on Vehicle Re-identification Methods with Unsupervised Learning
    XU Yan, GUO Xiaoyan, RONG Leilei
    Journal of Frontiers of Computer Science and Technology    2023, 17 (5): 1017-1037.   DOI: 10.3778/j.issn.1673-9418.2209100
    As one of the key technologies of intelligent transportation systems, vehicle re-identification (Re-ID) aims to retrieve the same vehicle from different monitoring scenes and plays an important role in building a safe and smart city. With the continuous development of computer vision, the Re-ID method of using supervised learning suffers from the problems of strong reliance on manual annotation in the training process and weak scene generalization ability, so unsupervised learning of vehicle Re-ID gradually becomes the focus of research in recent years. Firstly, the present mainstream vehicle Re-ID datasets and the commonly used model evaluation metrics are introduced. Then, latest unsupervised learning-based vehicle Re-ID methods are grouped into two categories: gene-rative adversarial networks and clustering algorithms according to the current research ideas. Starting from the problems of domain deviation, cross-view deviation and insufficient information of data samples, the former is further divided into three categories of style transfer, multi-view generation, and data augmentation. For the labeling pro-blem, the latter can be divided into two categories of pseudo-labeled unsupervised domain adaptation and no label infor-mation required. With problem solving as the starting point, the fundamentals, advantages and disadvantages, and performance results of each type of method on mainstream datasets are summarized. Finally, the challenges faced by the current unsupervised learning for vehicle Re-ID are analyzed, and the future work in this research direction is prospected.
    Reference | Related Articles | Metrics
    Research Progress of Graph Neural Network in Knowledge Graph Construction and Application
    XU Xinran, WANG Tengyu, LU Cai
    Journal of Frontiers of Computer Science and Technology    2023, 17 (10): 2278-2299.   DOI: 10.3778/j.issn.1673-9418.2302059
    As an effective representation of knowledge, knowledge graph network can be used to represent rich factual information between different categories and become an effective knowledge management tool. It has achieved great results in the application and research of knowledge engineering and artificial intelligence. Know-ledge graph is usually expressed as a complex network structure. Its unstructured characteristics make the applica-tion of graph neural network to the analysis and research of knowledge graph become a research hotspot in academia. The purpose of this paper is to provide extensive research on knowledge graph construction technology based on graph neural network to solve two types of knowledge graph construction tasks, including knowledge extraction (entity, relationship and attribute extraction) and knowledge merging and processing (link prediction, entity alignment and knowledge reasoning, etc.). Through these tasks, the structure of knowledge graph can be further improved and new knowledge and reasoning relationships can be discovered. This paper also studies the advanced graph neural network method for knowledge graph related applications, such as recommendation system, question answering system and computer vision. Finally, the future research directions of knowledge graph application based on graph neural network are proposed.
    Reference | Related Articles | Metrics
    Emotion Controllable Dialogue Generation Method Combining Fine-Tuning and Reranking
    DU Baoxiang, MA Zhiqiang, WANG Chunyu, JIA Wenchao, WANG Hongbin
    Journal of Frontiers of Computer Science and Technology    2023, 17 (4): 953-963.   DOI: 10.3778/j.issn.1673-9418.2106060
    The existing emotional dialogue generation methods are usually based on the sequence-to-sequence (Seq2Seq) dialogue generation models, and improve the encoding or decoding in terms of emotion. This method can achieve a certain extent of emotional response generation abilities but is prone to low-quality and common response problems. To solve these problems and achieve emotion controllable and high-quality response generation, this paper proposes an emotion controllable dialogue generation method combining fine-tuning and reranking, called EmoGPT (emotional generative pre-trained transformer). In the model training phase, a method for fine-tuning the GPT-2 model using dialogue corpus with emotion category labels is proposed to learn the semantic and emotional dependence of the sentences; in the response generation phase, an emotion reranking strategy is proposed to score and sort the generated multiple responses to improve the controllability of the response emotions. Experimental results of emotional response generation using a dialogue dataset with emotion labels show that EmoGPT with an emotion reranking strategy achieves better performance than comparison models in terms of content relevance and emotion consistency of the generated responses, thus verifying the ability of the proposed method to generate emotion controllable and high-quality responses.
    Reference | Related Articles | Metrics
    Low-Light Enhancement Method for Light Field Images by Fusing Multi-scale Features
    LI Mingyue, YAN Tao, JING Huahua, LIU Yuan
    Journal of Frontiers of Computer Science and Technology    2023, 17 (8): 1904-1916.   DOI: 10.3778/j.issn.1673-9418.2202064
    Light field images (LFI) record rich 3D structural and textural details of target scenes, which have great advantages in a wide range of computer vision tasks. However, LFI captured under low-light conditions always suffer from low brightness and strong noise, which may seriously degrade the quality of LFI. In this paper, a low-light LFI enhancement method by fusing multi-scale light field (LF) structural features is proposed , which adopts digital single-lens reflex camera (DSLR) images to supervise the generated enlightened LFI. To explore and exploit light field structural features, angular and spatial Transformers are introduced to extract LF structural features from LFI at different scales, i.e., the complementary information between sub-views and local and long-range dependencies within each sub-view. A recurrent fusion module is proposed to preserve the long-term memory of features at diffe-rent scales by using a long-short-term memory network, while adaptively aggregating LF structural features in the entire feature space through local and global fusion layers. A 4D residual reconstruction module is designed to reconstruct target LFI sub-views from the aggregated features. In additional, a dataset of low-light LFI and normal-light DSLR image pairs is constructed to train the proposed network. Extensive experiment demonstrates that the proposed network can effectively improve the quality of low-light LFI, and it obviously outperforms other state-of-the-art methods.
    Reference | Related Articles | Metrics
    Design and Implementation of Efficient Multi-branch Predictor
    YANG Ling, ZHOU Jinwen, WANG Jing, LAN Mengqiao, DING Zijian, YANG Shi, WANG Yongwen, HUANG Libo
    Journal of Frontiers of Computer Science and Technology    2023, 17 (8): 1842-1851.   DOI: 10.3778/j.issn.1673-9418.2207069
    Branch prediction is a momentous technology guarantee for processor performance, especially for the widely used superscalar processor. The properties of the branch predictor significantly affect the overall performance, power consumption, and area of the processor. To obtain a more cost-effective branch predictor in the superscalar processor, an attempt is made to use a single TAGE (tagged geometric history length) predictor to predict the branches within the fetch width. The championship branch prediction platform is used to evaluate the performance of the predictor, and its prediction ability is sufficient to meet the prediction conditions. However, in practice, conflicts in both the predictor and branch target buffer affect its performance. To solve the above problem, this paper adds additional prediction paths based on a single TAGE branch predictor and independently saves and predicts additional branch instruction information. This predictor is implemented in the processor using hardware description language and compared with a single TAGE branch predictor to perform standard benchmark programs for embedded processors, dhrystone, coremark and embench. Experimental results show that the performance of the optimized branch predictor is improved by 14.1 percentage points, while the storage overhead is only increased by 9.06%. Finally, through the analysis of the experimental data, it is found that this scheme is not only conducive to the prediction of additional branch instructions, but also can achieve more accurate prediction of single branch instruction through more accurate acquisition of branch history information.
    Reference | Related Articles | Metrics
    Survey of Deformable Convolutional Networks
    LIU Weiguang, LIU Dong, WANG Lu
    Journal of Frontiers of Computer Science and Technology    2023, 17 (7): 1549-1564.   DOI: 10.3778/j.issn.1673-9418.2209076
    In recent years, with the rapid development of deep learning, deformable convolutional networks have received extensive attention because of their powerful feature extraction capabilities, overcoming some problems that are difficult to solve in convolutional neural networks, and have played an important role in computer vision, natural language processing and other related fields. Since there is a little research on systematic summary of the deformable convolutional network, in order to provide a detailed reference for subsequent research, this paper summarizes the related work since the introduction of the deformable convolutional network. Firstly, this paper reviews the high-quality literature in recent years, and introduces the core technologies such as deformable convolution and deformable region of interest pooling in deformable convolutional networks from the perspective of invariant features. Secondly, the collected relevant literature is classified according to different research fields, and the appli-cation of deformable convolutional networks in image recognition and classification, target detection, image seg-mentation, target tracking and other research fields is comprehensively summarized. At the same time, the perfor-mance, advantages and disadvantages of important network models are listed. Thirdly, by combing the literature, the advantages and disadvantages of the deformable convolutional network are analyzed, and the possible future research trends of the deformable convolutional network are discussed according to some problems existing at the present stage. Finally, the deformable convolutional networks are summarized and prospected based on invariant feature extraction.
    Reference | Related Articles | Metrics
    Research on Blockchain Interoperability and Cross-Chain Technology
    WANG Qun, LI Fujuan, NI Xueli, XIA Lingling, LIANG Guangjun, MA Zhuo
    Journal of Frontiers of Computer Science and Technology    2023, 17 (8): 1749-1775.   DOI: 10.3778/j.issn.1673-9418.2212081
    Blockchain is a distributed ledger technology with multi-party consensus, traceability and tamper-proof, which provides a broad application prospect for constructing efficient, trusted and secure data sharing mechanism and optimizing business processes. However, when blockchain is in a stage of rapid development, how to realize cross-chain interaction of information and cross-chain transfer of value has become an urgent problem to be solved in the process of blockchain extension to depth. Firstly, based on the review of the existing research results, the concept of blockchain interoperability is proposed, and it is divided into five aspects: inter-chain interoperability, inter-layer interoperability, inter-fork interoperability, inter-slice interoperability and interoperability between on-chain and off-chain. Secondly, by sorting out the evolution and implementation of cross-chain operation of blockchain, and referring to the TCP/IP architecture, a cross-chain operation model is designed, and the main implementation steps are functionally described. Thirdly, in view of the current research status of cross-chain operation, four key technologies of cross-chain operation, including notary mechanism, side chain/relay, hash lock and distributed private key control, are selected for analysis. Then, combined with the technical characteristics and application scenarios, the application demonstration is highlighted, some typical cross-chain application projects are introduced, and the security of blockchain cross-chain operation is analyzed. Finally, the future development trend of blockchain interoperability and cross-chain technology is summarized and explored.
    Reference | Related Articles | Metrics
    Local and Global Feature Fusion Network Model for Aspect-Based Sentiment Analysis
    XIA Hongbin, LI Qiang, LIU Yuan
    Journal of Frontiers of Computer Science and Technology    2023, 17 (4): 902-911.   DOI: 10.3778/j.issn.1673-9418.2107069
    Aspect-based sentiment analysis aims to predict the sentiment polarity of a specific aspect in a sentence or document. Most of the recent research is to use attention mechanism and external semantic knowledge to model the global context. The sentiment polarity of an aspect often depends on the local context which is highly related to the aspect, but most models focus too much on the global context, which makes the parameter amount of the model generally larger, resulting in an increase in the amount of calculation. To this end, this paper proposes a lightweight network model based on multi-head attention mechanism, which is named local and global feature fusion network. Firstly, a bidirectional gated recurrent unit is used to encode the context. Secondly, context words that are less relevant to the aspect term are masked out according to the semantic relation distance with the aspect term, so as to obtain the local context representation. Finally, the local and global context are extracted respectively by multi-head Aspect-aware attention network, and the results of the two extraction are combined. In addition, the pre-trained BERT is also applied to the proposed model and better results are obtained. Experiments are conducted on three datasets: Twitter, Laptop, and Restaurant. Accuracy and F1 indicators are used for evaluation. Experimental results show that the proposed model achieves better results than other aspect-based sentiment classification algorithms with a small amount of parameters.
    Reference | Related Articles | Metrics
    Implementation and Application of Large-Scale Drug Virtual Screening
    ZHANG Baohua, LI Hui, LIU Qian, GAO Meina, HUANG He, ZHAO Yi, YU Kunqian, JIN Zhong
    Journal of Frontiers of Computer Science and Technology    2023, 17 (5): 1049-1056.   DOI: 10.3778/j.issn.1673-9418.2109074
    The molecular docking-based virtual screening technique evaluates the binding abilities between multiple ligand compounds and receptors to screen for the active compounds. In the context of the global spread of the COVID-19 pandemic, large-scale and rapid drug virtual screening is crucial for identifying potential drug molecules from massive datasets of ligand structures. The powerful computing power of supercomputer provides hardware guarantee for drug virtual screening, but the super large-scale drug virtual screening still faces many challenges that affects the effective execution of the calculation. Based on the analysis of the challenges, this paper proposes a centralized task distribution scheme with a central database, and designs a multi-level task distribution framework. The challenges are effectively solved through multi-level intelligent scheduling, multi-level compression processing of massive small molecule files, dynamic load balancing and high error tolerance management technology. An easy-to-use “tree” multi-level task distribution system is implemented. A fast, efficient and stable drug virtual screening task distribution, calculation and result analysis function is realized, and the computing efficiency is nearly linear. Then, heterogeneous computing technology is used to complete the drug virtual screening of more than 2 billion compounds, for two different active sites for COVID-19, on the domestic super computing system, which provides a powerful computing guarantee for the super large-scale rapid virtual screening of explosive malignant infectious diseases.
    Reference | Related Articles | Metrics
    Using HLS to Develop FPGA Heterogeneous Acceleration System: Problems, Optimization Methods and Opportunities
    XU Cheng, GUO Jinyang, LI Chao, WANG Jing, WANG Taolei, ZHAO Jieru
    Journal of Frontiers of Computer Science and Technology    2023, 17 (8): 1729-1748.   DOI: 10.3778/j.issn.1673-9418.2210102
    Currently, field programmable gate arrays (FPGAs) are favored by both academia and industry due to their programmability and excellent energy efficiency ratio. However, traditional FPGA development based on hardware description languages faces programming challenges. Hardware description languages, which are different from commonly used high-level languages, hinder software developers from utilizing FPGAs. High-level synthesis (HLS) enables developers to directly develop FPGA hardware from high-level languages such as C/C++, and is widely regarded as the preferred solution to this problem. In recent years, there have been many works in academia on HLS, dedicated to solving various problems in the HLS application process and improving the performance of systems developed through HLS. This paper lists feasible optimization directions from the perspective of heterogeneous system developers around the issue of developing FPGA heterogeneous systems using HLS. At the compilation optimization level, HLS tools can automatically generate high-performance RTL designs by inserting compilation guidance and designing efficient spatial exploration algorithms. At the memory access optimization level, HLS tools can set up buffers, split and replicate data to improve the overall system bandwidth. At the parallel optimization level, HLS tools can implement statement-level, task-level and board-level parallelism. Meanwhile, some technologies such as DSL, although they cannot directly improve the performance of heterogeneous acceleration systems, can further enhance the usability of HLS tools. Finally, this paper summarizes some challenges currently faced by HLS and prospects the future research on HLS.
    Reference | Related Articles | Metrics
    Review of Computing Offloading Schemes for Multi-access Edge Computing
    ZHANG Bingjie, YANG Yanhong, CAO Shaozhong
    Journal of Frontiers of Computer Science and Technology    2023, 17 (9): 2030-2046.   DOI: 10.3778/j.issn.1673-9418.2301068
    Under the background of the Internet of things, the development of massive and large-scale machine communication has brought about the explosive growth of data traffic. The traditional cloud computing model can no longer meet the needs of low delay and low energy consumption of terminal data processing. Multi-access edge computing (MEC) with distributed multi-nodes near the terminal side is becoming the best choice to solve this problem. Computational offloading is the key technology of MEC, the offloading performance is affected by many factors, and there is a large space for optimization. How to design a high-performance computational offloading scheme has become the main research goal of scholars at home and abroad. This paper reviews the research of computing offloading scheme for MEC, introduces the concept of MEC, sorts out the development and application of MEC, and the execution process of computing offloading, analyzes and compares the recent research methods of computing offloading. According to different improvements, a computing offloading scheme is summarized which takes the offloading system environment, offloading delay, energy consumption of mobile devices and multiple evaluation indexes as the optimization direction. The problems of resource allocation, universality and security in MEC-oriented computing offloading are presented, and the future research directions are forecasted based on these problems.
    Reference | Related Articles | Metrics
    Image Super-resolution Reconstruction Algorithm Based on Multi-scale Adaptive Upsampling
    LYU Jia, XU Pengcheng
    Journal of Frontiers of Computer Science and Technology    2023, 17 (4): 879-891.   DOI: 10.3778/j.issn.1673-9418.2107136
    In view of the problems in the existing convolutional neural network-based super-resolution network, such as difficult convergence of training, unable to adapt to multiple upsampling coefficients and unable to sample non-integer upsampling coefficients, an image super-resolution reconstruction algorithm of adaptive upsampling module based on multi-scale and the idea of divide-and-conquer is proposed in this paper. In the algorithm, the improved multi-scale channel attention feature extraction module is used to extract multi-scale features of low-resolution images to generate feature maps at different scales, and then the global feature fusion is realized by inputting them into the bottleneck layer. The super-resolution images are obtained by using the adaptive upsampling module based on divide-and-conquer to solve the adaptation problem of different upsampling coefficients and the upsampling problem of non-integer upsampling coefficients. In the contrast experiment, the proposed algorithm still has good convergence without any initialization methods. Under the integer magnification factor, the image reconstruction performance of the proposed algorithm exceeds the current mainstream super-resolution network, and the PSNR and SSIM performance are improved by 0.34 dB and 0.0391 respectively compared with MRFN. Under the non-integer magnification factor, the average PSNR performance is improved by 1.24 dB compared with the bicubic interpolation method, and there is not necessary to train each magnification factor.
    Reference | Related Articles | Metrics
    Deep Learning Compiler Load Balancing Optimization Method for Model Training
    WANG Li, GAO Kai, ZHAO Yaqian, LI Rengang, CAO Fang, GUO Zhenhua
    Journal of Frontiers of Computer Science and Technology    2024, 18 (1): 111-126.   DOI: 10.3778/j.issn.1673-9418.2209026
    For computing-intensive artificial intelligence (AI) training tasks, the computational graph is more complex, and data loading, task division of the computational graph, and load balancing of task scheduling have become the key factors affecting the computing performance. This paper proposes three optimization methods to make the task scheduling of model training in deep learning compilers reach the load balance state. Firstly, the load balance between CPU and back-end computing devices is realized by automatically establishing an efficient pipeline for data loading and model training, which improves the overall energy efficiency of the system. Secondly, the layered optimization technology of computational graph is used to realize the load balance of computational graph when the back-end devices are scheduling. Finally, this paper improves the resource utilization of back-end devices by automatically establishing efficient pipeline between layers. Experimental results show that the proposed optimization method achieves the system load balancing in the process of automatically mapping the training tasks to underlying hardware devices. Compared with traditional deep learning frameworks and compilers such as TensorFlow, nGraph, etc., this paper achieves 2%~10% performance improvement in the training of different AI models, and the overall power consumption of the training system can be reduced by more than 10%.
    Reference | Related Articles | Metrics