首页期刊导航|中国科学：信息科学（英文版）
期刊信息/Journal information
中国科学：信息科学（英文版）
中国科学：信息科学（英文版）

周光召

月刊

1674-733X

informatics@scichina.org

010-64015683

100717

北京东黄城根北街16号

中国科学：信息科学（英文版）/Journal Science China Information SciencesCSCDCSTPCDEISCI
查看更多>>《中国科学》是中国科学院主办、中国科学杂志社出版的自然科学专业性学术刊物。《中国科学》任务是反映中国自然科学各学科中的最新科研成果，以促进国内外的学术交流。《中国科学》以论文形式报道中国基础研究和应用研究方面具有创造性的、高水平的和有重要意义的科研成果。在国际学术界，《中国科学》作为代表中国最高水平的学术刊物也受到高度重视。国际上最具有权威的检索刊物SCI，多年来一直收录《中国科学》的论文。1999年《中国科学》夺得国家期刊奖的第一名。
正式出版
收录年代

    Categorizing methods for integrating machine learning with executable specifications

    David HARELRaz YERUSHALMIAssaf MARRONAchiya ELYASAF...
    1-15页
    查看更多>>摘要：Deep learning(DL),which includes deep reinforcement learning(DRL),holds great promise for carrying out real-world tasks that human minds seem to cope with quite readily.That promise is already delivering extremely impressive results in a variety of areas.However,while DL-enabled systems achieve excellent performance,they are far from perfect.It has been demonstrated,in several domains,that DL systems can err when they encounter cases they had not hitherto encountered.Furthermore,the opacity of the produced agents makes it difficult to explain their behavior and ensure that they adhere to various requirements posed by human engineers.At the other end of the software development spectrum of methods,behavioral programming(BP)facilitates orderly system development using self-standing executable modules aligned with how humans intuitively describe desired system behavior.In this paper,we elaborate on different approaches for combining DRL with BP and,more generally,machine learning(ML)with executable specifications(ES).We begin by defining a framework for studying the various approaches,which can also be used to study new emerging approaches not covered here.We then briefly review state-of-the-art approaches to integrating ML with ES,continue with a focus on DRL,and then present the merits of integrating ML with BP.We conclude with guidelines on how this categorization can be used in decision making in system development,and outline future research challenges.
      原文链接:
    • 万方数据
    • 维普

    Building a domain-specific compiler for emerging processors with a reusable approach

    Mingzhen LIYi LIUBangduo CHENHailong YANG...
    16-34页
    查看更多>>摘要：High-performance computing and deep learning domains have been motivating the design of domain-specific processors.Although these processors can provide promising computation capability,they are notorious for exotic programming paradigms.To improve programming productivity and fully exploit the performance potential of these processors,domain-specific compilers(DSCs)have been proposed.However,building DSCs for emerging processors requires tremendous engineering efforts because the commonly used compilation stack is difficult to be reused.Owing to the advent of multilevel intermediate representation(MLIR),DSC developers can leverage reusable infrastructure to extend their customized functionalities without rebuilding the entire compilation stack.In this paper,we further demonstrate the effectiveness of MLIR by extending its reusable infrastructure to embrace a heterogeneous many-core processor(Sunway processor).In particular,we design a new Sunway dialect and corresponding backend for the Sunway processor,fully exploiting its architectural advantage and hiding its programming complexity.To show the ease of building a DSC,we leverage the Sunway dialect and existing MLIR dialects to build a stencil compiler for the Sunway processor.The experimental results show that our stencil compiler,built with a reusable approach,can even perform better than state-of-the-art stencil compilers.
      原文链接:
    • 万方数据
    • 维普

    Learnware:small models do big

    Zhi-Hua ZHOUZhi-Hao TAN
    35-46页
    查看更多>>摘要：There are complaints about current machine learning techniques such as the requirement of a huge amount of training data and proficient training skills,the difficulty of continual learning,the risk of catastrophic forgetting,and the leaking of data privacy/proprietary.Most research efforts have been focusing on one of those concerned issues separately,paying less attention to the fact that most issues are entangled in practice.The prevailing big model paradigm,which has achieved impressive results in natural language processing and computer vision applications,has not yet addressed those issues,whereas becoming a serious source of carbon emissions.This article offers an overview of the learnware paradigm,which attempts to enable users not to need to build machine learning models from scratch,with the hope of reusing small models to do things even beyond their original purposes,where the key ingredient is the specification which enables a trained model to be adequately identified to reuse according to the requirement of future users who know nothing about the model in advance.
      原文链接:
    • 万方数据
    • 维普

    TULAM:trajectory-user linking via attention mechanism

    Hao LIShuyu CAOYaqing CHENMin ZHANG...
    47-64页
    查看更多>>摘要：Recently,the application of location-based services(LBS)has become a prevalent means to pro-vide convenience in customers'everyday lives.However,because massive volumes of location information are collected by LBS applications,users may suffer from serious privacy issues.Prior studies have shown that the identities of users can be inferred from historical anonymous trajectories,which is formulated as the trajectory-user linking(TUL)task.Although some recurrent neural network(RNN)-based models have been proposed to capture implicit movement patterns among trajectories to improve TUL performance,they cannot learn the sequential and contextual semantics within any individual trajectory completely,leaving the advantages of RNNs underutilized.We therefore propose an RNN model with an attention mechanism called TULAM to improve the accuracy of the TUL task.TULAM learns sequential relationships within individual trajectories via RNN and captures contextual semantics from trajectories via a multi-head atten-tion mechanism.Additionally,we propose a novel location encoding method called approximate one-hot to solve the corpus shortage problem of trajectory datasets.Evaluations were conducted on real datasets from the Gowalla and Foursquare LBS platforms.The experimental results indicate that TULAM is a practical solution that achieves significant improvements over existing methods with satisfactory model complexity and convergence.
      原文链接:
    • 万方数据
    • 维普

    HTDcr:a job execution framework for high-throughput computing on supercomputers

    Jiazhi JIANGDan HUANGHu CHENYutong LU...
    65-81页
    查看更多>>摘要：High-throughput computing(HTC)is a computing paradigm that aims to accomplish jobs by easily breaking them into smaller,independent components.However,it requires a large amount of computing power for a long time.Most existing HTC frameworks are job-oriented without support for coscheduling with hardware architecture and task-level execution.Also,most of the frameworks reach a limited scale,and their usability needs further improvement.Herein,we present HTDcr,a job execution framework for the HTC on supercomputers.This study aims to improve the throughput,task dispatching,and usability of the framework.In detail,the throughput optimizations include a sophisticated designed task management system,a hierarchical scheduler,and the co-optimization of the task-scheduling strategy with the application and hardware characteristics.The optimizations for usability include a programable execution workflow,mechanisms for more robust and reliable service qualities,and a fine-grained resource allocation system for the colocation of multiple jobs.According to our evaluations,HTDcr can achieve outstanding scalability and high throughput on large-scale clusters for the HTC workload.We evaluate HTDcr with several microbenchmarks and real-world applications on Tianhe-2 and Sunway TaihuLight to demonstrate its effects on existing design mechanisms.For instance,the task scheduling for two real-world applications integrated with the application and hardware characteristics achieves 1.7× and 1.9× speedups over the basic task-scheduling strategy.
      原文链接:
    • 万方数据
    • 维普

    ASTSDL:predicting the functionality of incomplete programming code via an AST-sequence-based deep learning model

    Yaoshen YUZhiqiu HUANGGuohua SHENWeiwei LI...
    82-100页
    查看更多>>摘要：Code recommendation systems have been widely used in helping developers implement unfamiliar programming tasks.Many existing code recommenders or code search engines can retrieve relevant code rapidly with high accuracy,however,they cannot recommend any code outside similar ones.We propose an approach to predict the functionality of incomplete programming code by using syntactical information,and providing a list of potential functionalities to guess what the developers want,in order to increase the diversity of recommendations.In this paper,we propose a deep learning model called ASTSDL,which uses a sequence-based representation of source code to predict functionality.We extract syntactical information from the abstract syntax tree(AST)of the source code,apply a deep learning model to capture the syntactic and sequential information,and predict the functionality of the source code fragments.The experimental results demonstrate that ASTSDL can effectively predict the functionality of incomplete code with an accuracy above 84％in the top-10 list,even if there is only half of the complete code.
      原文链接:
    • 万方数据
    • 维普

    Fewer is more:efficient object detection in large aerial images

    Xingxing XIEGong CHENGQingyang LIShicheng MIAO...
    101-119页
    查看更多>>摘要：Current mainstream object detection methods for large aerial images usually divide large images into patches and then exhaustively detect the objects of interest on all patches,no matter whether there exist objects or not.This paradigm,although effective,is inefficient because the detectors have to go through all patches,severely hindering the inference speed.This paper presents an objectness activation network(OAN)to help detectors focus on fewer patches but achieve more efficient inference and more accurate results,enabling a simple and effective solution to object detection in large images.In brief,OAN is a light fully-convolutional network for judging whether each patch contains objects or not,which can be easily integrated into many object detectors and jointly trained with them end-to-end.We extensively evaluate our OAN with five advanced detectors.Using OAN,all five detectors acquire more than 30.0％speed-up on three large-scale aerial image datasets,meanwhile with consistent accuracy improvements.On extremely large Gaofen-2 images(29200 x 27620 pixels),our OAN improves the detection speed by 70.5％.Moreover,we extend our OAN to driving-scene object detection and 4K video object detection,boosting the detection speed by 112.1％and 75.0％,respectively,without sacrificing the accuracy.
      原文链接:
    • 万方数据
    • 维普

    RGB oralscan video-based orthodontic treatment monitoring

    Yan TIANHanshi FUHao WANGYuqi LIU...
    120-135页
    查看更多>>摘要：Orthodontic treatment monitoring involves using current images and previous 3D models to estimate the relative position of individual teeth before and after orthodontic treatment.This process differs from image-based object 6D pose estimation due to the gingiva deformation and varying pose offsets for each tooth during treatment.Motivated by the fact that the poses of molars remain relatively fixed in implicit orthodontics,we design an approach that employs multiview pose evaluation and bidirectional temporal propagation for jaw pose estimation and then employs an iteration-based method for tooth alignment.To handle changes in tooth appearance or location with weak texture across frames,we also introduce an instance propagation module that leverages positional and semantic information to explore instance relations in the temporal domain.We evaluated the performance of our approach using both the Shining3D tooth pose dataset and the Aoralscan3 tooth registration dataset.Our experimental results demonstrate remarkable accuracy improvements compared with existing methods.
      原文链接:
    • 万方数据
    • 维普

    Prescribed-time leader-following consensus of linear multi-agent systems by bounded linear time-varying protocols

    Kai ZHANGBin ZHOUXuefei YANGGuangren DUAN...
    136-149页
    查看更多>>摘要：This paper considers the prescribed-time leader-following consensus problem of input-constrained linear multi-agent systems under generally directed communication topology in two cases:the Laplacian ma-trix related to the entire communication topology between agents is either known or unknown.In particular,the consensus problem for the former case is solved by a novel bounded linear time-varying(LTV)protocol,where the feedback gain is formulated by the parametric Lyapunov equation and the knowledge of the Lapla-cian matrix.Moreover,by utilizing a distributed observer,a fully bounded LTV protocol is proposed for the latter case.It should be noted that,compared with the existing results,the system under consideration is more general,the designed protocols are linear,and the consensus problem is accomplished even in a fully distributed manner.Finally,the effectiveness of the proposed approach is verified by a numerical example.
      原文链接:
    • 万方数据
    • 维普

    Error-based adaptive optimal tracking control of nonlinear discrete-time systems

    Chun LIJinliang DINGFrank L.LEWISTianyou CHAI...
    150-163页
    查看更多>>摘要：In this paper,for the output tracking problem of nonlinear discrete-time systems,a performance index is newly defined using the adaptive dynamic programming(ADP)technique to completely eliminate tracking errors in theory.In contrast to traditional definitions of performance indices in other ADP-based methods,the proposed performance index is not only designed from the perspective of output tracking errors but also introduced errors of system states and control inputs at adjacent stages,which is suitable for practical situations in many industrial applications,such as the alumina production,the flotation process,and the mineral grinding process.We proved that the obtained controller can make the system output fully track the given reference trajectory by applying the iterative criteria of the ADP technique.In addition,the proposed algorithm was implemented using a data-driven technique and neural networks to avoid analyzing and deducing the complicated dynamics of actual industrial processes.Finally,using historical data for a forced-circulation evaporation system in the alumina production process,the effect of the proposed approach was verified through a numerical simulation and compared with that of the proportional-integral controller.
      原文链接:
    • 万方数据
    • 维普