检索结果-南通市图书馆

SWG: an architecture for sparse weight gradient computation

Science China(Information Sciences) 2024年第2期67卷 302-321页

作者： Weiwei WU Fengbin TU Xiangyu LI Shaojun WEI Shouyi YIN School of Integrated Circuits Tsinghua University Department of Electronic and Computer Engineering The Hong Kong University of Science and Technology

On-device training for deep neural networks(DNN) has become a trend due to various user preferences and scenarios. The DNN training process consists of three phases, feedforward(FF), backpropagation(BP), and weight gradient(WG) update. WG takes about one-third of the computation in the whole training process. Current training accelerators usually ignore the special computation property of WG and process it in a way similar to FF/BP. Besides, the extensive data sparsity existing in WG, which brings opportunities to save computation, is not well explored. Nevertheless, exploiting the optimization opportunities would meet three underutilization problems, which are caused by(1) the mismatch between WG data dimensions and hardware parallelism,(2) the full sparsity, i.e., the sparsity of feature map(Fmap),error map(Emap), and gradient, and(3) the workload imbalance resulting from irregular sparsity. In this paper, we propose a specific architecture for sparse weight gradient(SWG) computation. The architecture is designed based on hierarchical unrolling and sparsity-aware(HUSA) dataflow to exploit the optimization opportunities of the special computation property and full data sparsity. In HUSA dataflow, the data dimensions are unrolled hierarchically on the hardware architecture. A valid-data trace(VDT) mechanism is embedded in the dataflow to avoid the underutilization caused by the two-sided input sparsity. The gradient is unrolled in PE to alleviate the underutilization induced by output sparsity while maintaining the data reuse opportunities. Besides, we design an intra-and inter-column balancer(IIBLC) to dynamically tackle the workload imbalance problem resulting from the irregular sparsity. Experimental results show that with HUSA dataflow exploiting the full sparsity, SWG achieves a speedup of 12.23× over state-of-the-art gradient computation architecture, Train Ware. SWG helps to improve the energy efficiency of the state-of-the-art training accelerator LNPU from 7.56 to 10.58 TOPS/W.

关键词： CNN training gradient computation sparsity architecture

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Channel Estimation for Reconfigurable Intelligent Surface Aided Multiuser Millimeter-Wave/THz Systems

引用

China Communications 2024年第3期21卷 91-103页

作者： Chu Hongyun Pan Xue Li Baijiang School of Communications and Information Engineering Xi’an University of Posts and TelecommunicationsXi’an 710121China

It is assumed that reconfigurable intelligent surface(RIS)is a key technology to enable the potential of mmWave communications.The passivity of the RIS makes channel estimation difficult because the channel can only be measured at the transceiver and not at the RIS.In this paper,we propose a novel separate channel estimator via exploiting the cascaded sparsity in the continuously valued angular domain of the cascaded channel for the RIS-enabled millimeter-wave/Tera-Hz systems,i.e.,the two-stage estimation method where the cascaded channel is separated into the base station(BS)-RIS and the RIS-user(UE)ones.Specifically,we first reveal the cascaded sparsity,i.e.,the sparsity exists in the hybrid angular domains of BS-RIS and the RIS-UEs separated channels,to construct the specific sparsity structure for RIS enabled multi-user systems.Then,we formulate the channel estimation problem using atomic norm minimization(ANM)to enhance the proposed sparsity structure in the continuous angular domains,where a low-complexity channel estimator via Alternating Direction Method of Multipliers(ADMM)is proposed.Simulation findings demonstrate that the proposed channel estimator outperforms the current state-of-the-arts in terms of performance.

关键词： atomic norm minimization cascaded channel estimation convex optimization mmWave/THz reconfigurable intelligent surface(RIS) sparsity

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Binary Random Projections with Controllable sparsity Patterns

引用

Journal of the Operations Research Society of China 2022年第3期10卷 507-528页

作者： Wen-Ye Li Shu-Zhong Zhang The Chinese University of Hong Kong Shenzhen 518172GuangdongChina Shenzhen Research Institute of Big Data Shenzhen 518052GuangdongChina University of Minnesota MinneapolisMN 55455USA

Random projection is often used to project higher-dimensional vectors onto a lower-dimensional space,while approximately preserving their pairwise distances.It has emerged as a powerful tool in various data processing tasks and has attracted considerable research interest.Partly motivated by the recent discoveries in neuroscience,in this paper we study the problem of random projection using binary matrices with controllable sparsity patterns.Specifically,we proposed two sparse binary projection models that work on general data vectors.Compared with the conventional random projection models with dense projection matrices,our proposed models enjoy significant computational advantages due to their sparsity structure,as well as improved accuracies in empirical evaluations.

关键词： Binary random projection sparsity Dimensionality

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Improving time–frequency sparsity for enhanced audio source separation in degenerate unmixing estimation technique algorithm

引用

Journal of Control and Decision 2022年第4期9卷 502-515页

作者： Shahin M.Abdulla J.Jayakumari Department of Electronics and Communication Engineering Noorul Islam Centre for Higher EducationNagercoilIndia Department of Electronics and Communication Engineering Mar Baselios College of Engineering and TechnologyThiruvananthapuramIndia

In recent years,much research has been focused on separating acoustic sources from their mixtures.Degenerate Unmixing Estimation Technique(DUET)is one of the widely popular meth-ods of Blind Source Separation(BSS)in underdetermined scenarios.DUET is based on a signal recovery sparsity algorithm whose performance is strongly influenced by sparsity in the Time-Frequency(TF)domain.Noises and an several sources in mixtures limit the sparsity resulting in performance degradation in DUET.Here an enhanced strategy has been adopted by combin-ing DUET with adaptive noise cancellation utilising the Dual-Tree Complex Wavelet Transform(DTCWT)as a pre-processor and TF refinement utilising Synchroextracting Transform(SET)as a post-processor.This improves the sparsity of sources and energy concentrations in a TF rep-resentation.Results of the signal separation performance evaluation reveal that the proposed algorithm outperforms conventional DUET in signal separation,especially in real-time scenarios.

关键词： Blind source separation time-frequency dual-tree complex wavelet transform synchroextracting sparsity

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Ordinal factorization machine with hierarchical sparsity

引用

Frontiers of Computer Science 2020年第1期14卷 67-83页

作者： Shaocheng GUO Songcan CHEN Qing TIAN College of Computer Science and Technology Nanjing University of Aeronautics and AstronauticsNanjing 210000China School of Computer and Software Nanjing University of Information Science and TechnologyNanjing 210044China

Ordinal regression(OR)or classification is a machine learning paradigm for ordinal labels.To date,there have been a variety of methods proposed including kernel based and neural network based methods with significant performance.However,existing OR methods rarely consider latent structures of given data,particularly the interaction among covariates,thus losing interpretability to some extent.To compensate this,in this paper,we present a new OR method:ordinal factorization machine with hierarchical sparsity(OFMHS),which combines factorization machine and hierarchical sparsity together to explore the hierarchical structure behind the input variables.For the sake of optimization,we formulate OFMHS as a convex optimization problem and solve it by adopting the efficient alternating directions method of multipliers(ADMM)algorithm.Experimental results on synthetic and real datasets demonstrate the superiority of our method in both performance and significant variable selection.

关键词： ordinal regression factorization machine hierarchical sparsity interaction modelling

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Influence of the sparsity of random speckle illumination on ghost imaging in a noise environment

引用

Chinese Optics Letters 2021年第4期19卷 17-21页

作者：潘龙邓陈进于翠萍岳帅龚文林韩申生 Key Laboratory for Quantum Optics Shanghai Institute of Optics and Fine MechanicsChinese Academy of SciencesShanghai 201800China University of Chinese Academy of Sciences Beijing 100049China Wuhan Optics Valley Aerospace Sanjiang Laser Industrial Technology Research Institute Co. Ltd.Wuhan 430075China Hangzhou Institute for Advanced Study University of Chinese Academy of SciencesHangzhou 310024China

The influence of the sparsity of random speckle illumination on traditional ghost imaging(GI) and GI via sparsity constraint(GISC) in a noise environment is investigated. The experiments demonstrate that both GI and GISC obtain their best imaging quality when the sparsity of random speckle illumination is 0.5, which is also explained by some parameters such as detection of the signal to noise ratio and mutual coherence of the measurement matrix.

关键词： ghost imaging sparsity speckle illumination noise environment

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Three-Dimensional Interferometric ISAR Sensors Imaging for the Ship Target with Two-Dimensional sparsity

引用

Journal of Harbin Institute of Technology(New Series) 2019年第2期26卷 19-31页

作者： Yong Wang Xuefei Chen Research Institute of Electronic Engineering Technology Harbin Institute of Technology Key Laboratory of Marine Environmental Monitoring and Information Processing Ministry of Industry and Information Technology

There are great challenges for traditional three-dimensional( 3-D) interferometric inverse synthetic aperture radar( In ISAR) imaging algorithms of ship targets w ith 2-D sparsity in actual radar imaging system. To deal w ith this problem,a novel 3-D In ISAR imaging method is proposed in this paper.First,the high-precision gradient adaptive algorithm w as adopted to reconstruct the echoes in range dimension. Then the method of minimizing the entropy of the average range profile w as applied to estimate the parameters w hich are used to compensate translation components of the received echoes. Besides,the phase adjustment and image coregistration of the sparse echoes w ere achieved at the same time through the approach of the joint phase autofocus. Finally,the 3-D geometry coordinates of the ship target w ith 2-D sparsity w ere reconstructed by combining the range measurement and interferometric processing of the ISAR images. Simulation experiments w ere carried out to verify the practicability and effectiveness of the algorithm in the case that the received echoes are in 2-D sparsity.

关键词： 3-D InISAR 2-D sparsity gradient adaptive algorithm average range profile joint phase autofocus

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Day-ahead Network-constrained Unit Commitment Considering Distributional Robustness and Intraday Discreteness:A Sparse Solution Approach

引用

Journal of Modern Power Systems and Clean Energy 2023年第2期11卷 489-501页

作者： Xiaodong Zheng Baorong Zhou Xiuli Wang Bo Zeng Jizhong Zhu Haoyong Chen Waisheng Zheng School of Electrical Engineering Xi’an Jiaotong UniversityXi’an 710049China Electric Power Research Institute of China Southern Power Grid Co. Ltd.Guangzhou 510663China Shaanxi Key Laboratory of Smart Grid School of Electrical EngineeringXi’an Jiaotong UniversityXi’an 710049China Department of Industrial Engineering and the Department of Electrical and Computer EngineeringUniversity of PittsburghPittsburghPA 15260USA School of Electric Power Engineering South China University of TechnologyGuangzhou 510641China China Southern Power Grid Co. Ltd.Guangzhou 510663China

Quick-start generation units are critical devices and flexible resources to ensure a high penetration level of renewable energy in power systems.By considering the wind uncertainty and both binary and continuous decisions of quickstart generation units within the intraday dispatch,we develop a Wasserstein-metric-based distributionally robust optimization model for the day-ahead network-constrained unit commitment(NCUC)problem with mixed-integer recourse.We propose two feasible frameworks for solving the optimization problem.One approximates the continuous support of random wind power with a finite number of events,and the other leverages the extremal distributions instead.Both solution frameworks rely on the classic nested column-and-constraint generation(C&CG)method.It is shown that due to the sparsity of L_(1)-norm Wasserstein metric,the continuous support of wind power generation could be represented by a discrete one with a small number of events,and the rendered extremal distributions are sparse as well.With this reduction,the distributionally robust NCUC model with complicated mixed-integer recourse problems can be efficiently handled by both solution frameworks.Numerical studies are carried out,demonstrating that the model considering quick-start generation units ensures unit commitment(UC)schedules to be more robust and cost-effective,and the distributionally robust optimization method captures the wind uncertainty well in terms of out-of-sample tests.

关键词： Unit commitment distributional robustness mixed-integer recourse nested column-and-constraint generation(C&CG) sparsity Wasserstein metric

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

Low-Density Parity-Check Codes:Highway to Channel Capacity

引用

China Communications 2023年第2期20卷 235-256页

作者： Liyuan Song Shuyan Yu Qin Huang Fujian Provincial Engineering Technology Research Center of Photoelectric Sensing Application Key Laboratory of OptoElectronic Science and Technology for Medicine of Ministry of EducationFujian Normal UniversityFuzhou 350007China School of Electronic and Information Engineering Beihang UniversityBeijing 100191China

Low-density parity-check(LDPC)codes are not only capacity-approaching,but also greatly suitable for high-throughput implementation.Thus,they are the most popular codes for high-speed data transmission in the past two decades.Thanks to the low-density property of their parity-check matrices,the optimal maximum a posteriori probability decoding of LDPC codes can be approximated by message-passing decoding with linear complexity and highly parallel nature.Then,it reveals that the approximation has to carry on Tanner graphs without short cycles and small trapping sets.Last,it demonstrates that well-designed LDPC codes with the aid of computer simulation and asymptotic analysis tools are able to approach the channel capacity.Moreover,quasi-cyclic(QC)structure is introduced to significantly facilitate their high-throughput implementation.In fact,compared to the other capacity-approaching codes,QC-LDPC codes can provide better area-efficiency and energy-efficiency.As a result,they are widely applied in numerous communication systems,e.g.,Landsat satellites,Chang’e Chinese Lunar mission,5G mobile communications and so on.What’s more,its extension to non-binary Galois fields has been adopted as the channel coding scheme for BeiDou navigation satellite system.

关键词： LDPC codes sparsity high-speed message-passing cycles trapping sets quasi-cyclic

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

A supervised learning detection method with pre-processing of sparsity-based adaptive line enhancer

引用

Chinese Journal of Acoustics 2021年第4期40卷 496-510页

作者： JIN Shenglong CHI Cheng LI Yu HUANG Haining Institute of Acoustics Chinese Academy of SciencesBeijing 100190 Key Laboratory of Science and Technology on Advanced Underwater Acoustic Signal Processing Chinese Academy of SciencesBeijing 100190 University of Chinese Academy of Sciences Beijing 100049

For the weak line-spectrum target detection of uniiianned underwater vehicles in the complex environment,a supervised learning detection method witli pre-processing of sparsity-based adaptive line enhancer(ALE)is proposed.This method incorporates al_(p)-nonn sparse penalty into the cost function of ALE,and it also promotes the sparse regularization model to the 0sparsity-based ALE,the entropy features of target beam spectrum become obviously different.Using the small sample learning ability of support ve'etor machine(SVM),the method classifies the entropy characteristic curve of beam spectrum and determines if the target exists.The simulation result sliows that with the-20 dB input SNR,the SNR gain of l_(1/2)-nonn sparsity-based ALE is 11.5 dB higher than that of conventional ALE.The effectiveness of the method is verified by using the unmanned underwater vehicle(UUV)experimental data.Under the influence of wideband strong interferences,the false alarm rate is 3.5%and the detection rate is 95.8%,which improved the detection probability of weak line-spectrum targets.

关键词： sparsity underwater spectrum

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

欢迎您,

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

请选择保存的检索档案：

请选择收藏分类：

通借通还

欢迎您,

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：