检索结果-南通市图书馆

一种基于自适应PoT量化的无乘法神经网络训练方法

在线全文

学校读者我要写书评

暂无评论

高技术通讯 2024年第6期34卷 567-577页

作者：刘畅张蕊支天中国科学院大学北京100049 中国科学院计算技术研究所智能处理器研究中心北京100190 中国科学院计算技术研究所处理器芯片全国重点实验室北京100190

当前的深度神经网络的训练过程中需要包含大量的全精度乘累加(MAC)操作,导致神经网络模型的线性层(包含卷积层和全连接层)的计算过程所需的能耗占整体能耗的绝大部分,达90%以上。本文提出了一种自适应逐层缩放的量化训练方法,可支持在... 详细信息

当前的深度神经网络的训练过程中需要包含大量的全精度乘累加(MAC)操作,导致神经网络模型的线性层(包含卷积层和全连接层)的计算过程所需的能耗占整体能耗的绝大部分,达90%以上。本文提出了一种自适应逐层缩放的量化训练方法,可支持在神经网络计算全流程(前向传播和后向传播)将全部线性层中的全精度乘法替换为4位定点数加法计算和1位异或运算。实验结果表明,上述方法在能耗和准确率方面都优于现有方法,可支撑在训练过程中减少达95.8%的线性层能耗,在ImageNet数据集上的卷积神经网络和在WMT En-De任务上的Transformer网络得到小于1%的精度损失。

关键词：神经网络量化训练加速低能耗

Information-Theoretic Analysis for the Difficulty of Extracting Hidden Information

在线全文

学校读者我要写书评

暂无评论

Information-Theoretic Analysis for the Difficulty of Extract...

Conceptual design of the HIRFL-CSR external-target experiment

The 1st chinese Conference on Trusted computing and Information Security

作者： zhang Wei-ming, LI Shi-qu, CAO Jia, LIU Jiu-fen Department of Information Research, Information Engineering University,Zhengzhou 450002, Henan, china State Key Laboratory of Information Security, (Graduate School of the chinese academy sciences), beijing 100039, china

＜正＞ The difficulty of extracting hidden information, which is essentially a kind of secrecy, is analyzed by information-theoretic method. The relations between key rate, message rate, hiding capacity and difficulty of extraction are studied in the terms of unicity distance of stego-key, and the theoretic conclusion is used to analyze the actual extracting attack on Least Significant Bit（LSB） steganographic algorithms.

关键词： information hiding steganography extracting attack difficulty of extraction unicity distance

来源： cnki会议评论

在线全文

cnki会议

学校读者我要写书评

暂无评论

Science china(Physics,Mechanics & Astronomy) 2017年第1期60卷 65-71页

作者： LiMing Lu Hart Yi ZhiGang Xiao Ming Shao Song zhang guoqing Xiao Nu Xu Department of Physics Tsinghua University Beijing 100084 China Collaborative Innovation Center of Quantum Matter Beijing 100084 China Department of Modern Physics University of Science and Technology of China Hefei 230026 China Shanghai institute of Applied Physics Chinese Academy of Sciences Shanghai 201800 China institute of Modern Physics Chinese Academy of Sciences Lanzhou 730000 China institute of Particle Physics Central China Normal University Wuhan 430079 China

One of the main purposes of heavy-ion collisions over a wide range of beam energy is to study the bulk properties of strong interaction matter and understand the Quantum Chromo Dy- namics （QCD） phase diagram, which carries wealth of infor- mation of the phase transition and the possibly existing criti- cal point of the strongly interacting system [1]. Such system exists as hadron gases at lower temperature and low baryon density. By increasing the temperature or density, the bound- ary of the hadrons disappears and the confined quarks move freely in the whole system.

关键词： MRPC Conceptual design of the HIRFL-CSR external-target experiment HIRFL TPC CSR

在线全文

学校读者我要写书评

暂无评论

火-储耦合系统深度调峰综合经济性分析

储能科学与技术 2024年第10期13卷 3693-3705页

作者：张成凤朱轶林胡东子富征阳徐玉杰沈国清王亮陈海生华北电力大学能源动力与机械工程学院北京102206 中国科学院工程热物理研究所北京100190 华北电力大学能源动力与机械工程学院河北保定071003

火电机组耦合储热技术,可提高机组的热电解耦能力,减少深度调峰对系统安全性和经济性的影响。本工作提出了火电机组与填充床储热的耦合系统,在考虑了机组变工况、填充床储/释热过程动态时间序列基础上,通过EBSILON建立了耦合系统变工况... 详细信息

火电机组耦合储热技术,可提高机组的热电解耦能力,减少深度调峰对系统安全性和经济性的影响。本工作提出了火电机组与填充床储热的耦合系统,在考虑了机组变工况、填充床储/释热过程动态时间序列基础上,通过EBSILON建立了耦合系统变工况仿真模型;分析深度调峰对耦合系统热力性和碳排放量的影响,通过汽水分离器筒体应力变化分析深调对锅炉寿命损耗的影响,通过转子寿命损耗率曲线分析深调对汽轮机寿命的影响,最终建立耦合系统调度运行经济性模型,开展综合经济性分析。结果表明:火-储耦合系统比自身变工况减小碳排放量7418 t/a(仅配置风电)~9216 t/a(仅配置光伏);深度调峰对锅炉寿命损耗的影响大于汽轮机,火-储耦合系统可提高系统运行寿命,深度调峰318次/年,相比于机组自身变工况(30%~20%额定负荷)可提高13.3%~15.3%;火-储耦合系统深度调峰收益高于火电机组自身变工况,当填充床释热量用于发电或供热时,耦合系统比自身变工况收益分别增加40万元/年、72万元/年。

关键词：火电机组储热深度调峰综合经济性系统寿命碳排放

维普期刊数据库博看期刊

The High Energy X-ray telescope (HE) onboard the Insight-HXMT astronomy satellite

在线全文

学校读者我要写书评

暂无评论

Science china(Physics,Mechanics & Astronomy) 2020年第4期63卷 20-34页

作者： CongZhan Liu YiFei zhang XuFang Li XueFeng Lu Zhi Chang ZhengWei Li AiMei zhang YongJie Jin HuiMing Yu Zhao zhang MinXue Fu YiBao Chen JianFeng Ji YuPeng Xu JingKang Deng RenCheng Shang guoqing Liu FangJun Lu ShuangNan zhang YongWei Dong TiPei Li Mei Wu YanGuo Li HuanYu Wang BoBing Wu YongJie zhang Zhi zhang ShaoLin Xiong Yuan Liu Shu zhang HongWei Liu YiJung Yang Fan zhang Key Laboratory of Particle Astrophysics Institute of High Energy PhysicsChinese Academy of SciencesBeijing 100049China Department of Engineering Physics Tsinghua UniversityBeijing 100084China Department of Physics Tsinghua UniversityBeijing 100084China University of chinese academy of sciences Beijing 100049China

The Insight-Hard X-ray Modulation Telescope(Insight-HXMT) is a broadband X-ray and γ-ray(1-3000 ke V) astronomy satellite. One of its three main telescopes is the High Energy X-ray telescope(HE). The main detector plane of HE comprises 18 Na I(Tl)/Cs I(Na) phoswich detectors, where Na I(Tl) is used as the primary detector to measure ~ 20-250 ke V photons incident from the field of view(FOV) defined by collimators, and Cs I(Na) is used as the active shielding detector to Na I(Tl) by pulse shape discrimination. Additionally, Cs I(Na) is used as an omnidirectional γ-ray monitor. The HE collimators have a diverse FOV,i.e. 1.1°×5.7°(15 units), 5.7°×5.7°(2 units), and blocked(1 unit). Therefore, the combined FOV of HE is approximately5.7°×5.7°. Each HE detector has a diameter of 190 mm resulting in a total geometrical area of approximately 5100 cm2, and the energy resolution is ~15% at 60 ke V. For each recorded X-ray event by HE, the timing accuracy is less than 10 μs and the deadtime is less than 10 μs. HE is used for observing spectra and temporal variability of X-ray sources in the 20-250 ke V band either by pointing observations for known sources or scanning observations to unveil new sources. Additionally, HE is used for monitoring the γ-ray burst in 0.2-3 Me V band. This paper not only presents the design and performance of HE instruments but also reports results of the on-ground calibration experiments.

关键词： X-ray telescope HXMT calibration scintillation detector

Mapping of a new gene for brown planthopper resistance in cultivated rice introgressed from Oryza eichingeri

在线全文

学校读者我要写书评

暂无评论

chinese Science Bulletin 2001年第17期46卷 1459-1462页

作者： LIU guoqing YAN Huihuang FU Qiang QIAN Qian zhang Zhitao ZHAI Wenxue ZHU Lihuang institute of Genetics Chinese Academy of SciencesBeijing 100101China china National Rice Research institute Hangzhou 310006China

Wild rice species is an important source of useful genes for cultivated rice improvement Some accessions of Oryza eichingeri (In = 24, CC) from Africa confer strong resistance to brown planthopper (BPH), whitebacked planthopper (WBPH) and bacterial blight (KB). In the present study, restriction fragments length polymorphism (RFLP) and simple sequence repeats (SSR) analysis were performed on disomic backcross plants between Oryza saliva (2n =24, AA) and O. eichingeri in order to identify the presence of O, eichingeri segments and further to localize BPH-resistant gene. In the introgression lines, 1-6 O. eichingeri segments were detected on rice chromosomes 1, 2, 6, or/and 10. The dominant BPH resistant gene, tentatively named Bphl3(t), was mapped to chromosome 2, being 6.1 and 5.5 cM away from two microsatellite markers RM240 and RM250, respectively. The transfer and localization of this gene from O. eichingeri will contribute to the improvement of BPH resistance in cultivated rice.

关键词： cultivated rice Oryza eichingeri brown planthopper resistance gene mapping.

Wavelet-Based Denoising Attack on Image Watermarking

在线全文

学校读者我要写书评

暂无评论

Wavelet-Based Denoising Attack on Image Watermarking

基于模型定义的光学遥感卫星星上处理系统设计

The 1st chinese Conference on Trusted computing and Information Security

作者： XUAN Jian-hui, WANG Li-na, zhang Huan-guo School of Computer. Wuhan University, Wuhan 430079, Hubei, china National Laboratory of Pattern Recognition, chinese academy of Science, beijing 100080,china

＜正＞ In this paper,we propose wavelet-based denoisingattack methods on image watermarking in discrete cosine transform （DCT） or discrete Fourier transform （DFT） domain or discrete wavelet transform （DWT） domain. Wiener filtering based on wavelet transform is performed in approximationsubband to remove DCT or DFT domain watermark, and adaptive wavelet soft thresholding is emploved to remove the watermark resided in detail subbands of DWT domain.

关键词： watermarking watermarking attack wavelet denoising Wiener filter

来源： cnki会议评论

在线全文

cnki会议

学校读者我要写书评

暂无评论

测绘学报 2024年第4期53卷 689-699页

作者：刘薇刘松林郭子博刘凯张荔哲智慧地球重点实验室北京100029 西安测绘研究所陕西西安710054 西安电子科技大学陕西西安710071 中国空间技术研究院西安分院陕西西安710000

本文提出了一种基于模型定义的光学遥感卫星星上处理系统设计方法,构建了“硬件资源-算子模块-通用处理核-典型应用”的星上处理任务行为模型范式。在硬件层面,采用星上异构嵌入式计算平台,将多处理器一体化设计,通过标准化高速数据互连... 详细信息

本文提出了一种基于模型定义的光学遥感卫星星上处理系统设计方法,构建了“硬件资源-算子模块-通用处理核-典型应用”的星上处理任务行为模型范式。在硬件层面,采用星上异构嵌入式计算平台,将多处理器一体化设计,通过标准化高速数据互连,完成高速信号传输和数据处理;通过网络拓扑实现良好的扩展性,支持设备规模及数据处理复杂性的变化;在软件层面,针对星上智能处理任务中频繁数据读写操作、大量重复计算操作、卷积神经网络的通用计算与加速等需求,设计了星上平台指令集、星上算法通用算子和星上智能网络组件组成的可配置算子模块,并基于该模块可快速实现特定算法的硬件IP核。经仿真试验验证,本文方法可根据卫星平台和星上处理任务需求,按需适配最佳软硬件解决方案,并有效提高了计算资源利用率;提出的结合云检测的实时流水压缩编码方案,显著提升了压缩性能;设计的轻量化目标检测识别方法,计算资源效率达到91.5%;以高分一号、高分七号原始数据率为例进行分析,相比常规方案整体计算资源利用率分别提高了16.51%、17.77%。实现了星上处理系统研制过程模型化、工作模式可定义、逻辑资源共享化,是适应卫星小型化、快速部署的一种合理优化选择。

关键词：遥感卫星星上处理模型定义按需适配异构计算

维普期刊数据库博看期刊

在线全文

学校读者我要写书评

暂无评论

5G公网信号上高速铁路1+N组网方案探讨

铁道运输与经济 2024年第4期46卷 94-100页

作者：张伟辉张轩费汉明贾建刚王富章中国铁道科学研究院研究生部北京100081 中国铁道科学研究院集团有限公司国铁吉讯科技有限公司北京100081

为解决在高速动车组车厢内5G信号覆盖差、不能为旅客提供优质5G服务的问题,结合旅客手机归属于不同运营商的现状和高速铁路轨旁有多家运营商网络覆盖的实际情况,提出一种全新5G公网信号上高速铁路的组网方案。方案从车厢内网络、车地回... 详细信息

为解决在高速动车组车厢内5G信号覆盖差、不能为旅客提供优质5G服务的问题,结合旅客手机归属于不同运营商的现状和高速铁路轨旁有多家运营商网络覆盖的实际情况,提出一种全新5G公网信号上高速铁路的组网方案。方案从车厢内网络、车地回传网络以及地面网络等多环节的联合组网出发展开研究,重点论述车地回传中轨旁运营商网络通道的选择,从技术可行性、网络复杂度、社会集约化等多因素考虑,结合带宽叠加、智能选路以及边缘计算等技术,探讨实现5G公网信号上高速铁路的最优方案。此外,分析了旅客在车厢内通过5G信号访问互联网前注册5G核心网关键流程。最后研究了该组网方案下的各种经济效益和社会效益,全面完善方案的可落地性,为5G公网信号上高速铁路的部署提供参考。

关键词： 5G 小基站边缘计算多通路回传共建共享