检索结果-内蒙古大学图书馆

IEEE International Conference on Big Data and Cloud Computing (BdCloud)

作者： Lin Chen Kang He Hao Jiang Roberto Barrio Jie Liu Tiejun Li Science and Technology on Parallel and Distributed Processing Laboratory Laboratory of Software Engineering for Complex Systems National University of Defense Technology Changsha China College of Computer Science and Technology National University of Defense Technology Changsha China University of Zaragoza Zaragoza Spain

Quantization is an effective method for compressing Deep Neural Networks. Now, it is considered to accelerate the traditional HPC applications. In this article, we present a quantization method (q8gemm) using integer 8-bits(INT8) for matrix multiplication, which is implemented by block techniques to overcome the overflow problem. We choose different scales for one matrix’s each row and those for the other matrix’s each column to keep the numbers in range. With the intel advanced matrix extension cblas_gemm_s8u8s32, our q8gemm can accelerate panel-matrix multiplication in LU factorization. Furthermore, we propose a new mixed-precision solver (q8gesv 1 ) for the linear equations. Using q8gemm in INT8, q8gesv modifies LU factorization in FP32 and then utilizes GMRES iterative refinement in FP64 to bring the solution back to 64 bits accuracy. Numerical tests show that on two Intel CPU platforms with avx2 and avx512, q8gemm achieves a speedup of 1.313-1.621 over cblas_sgemm of MKL. The execution time of cblas_gemm_s8u8s32 makes up 86.78%-90.85% of the q8gemm. By means of q8gemm, our q8gesv can achieve a speedup of 1.176-1.493 over the optimized Higham’s mixed-precision algorithm utilized in the HPL-AI benchmark and a speedup of 1.1-1.459 over the LAPACKE_dsgesv function from MKL LAPACK. Meanwhile, q8gesv runs up to 2.859X faster on Intel(R) Core(TM) i5-10600KF and to 2.224X faster on Intel(R) Xeon(R) Gold 6130 than the LAPACKE_dgesv function from MKL LAPACK.

关键词： Deep learning Gold Quantization (signal) Benchmark testing

来源：评论

学校读者我要写书评

暂无评论

An Effective Design to Improve the Efficiency of DPUs on FPGA

An Effective Design to Improve the Efficiency of DPUs on FPG...

引用

International Conference on parallel and distributed Systems (ICPADS)

作者： Yutian Lei Qingyong Deng Saiqin Long Shaohui Liu Sangyoon Oh The Key Laboratory of Hunan Province for Internet of Things and Information Security The School of Computer Science Science and Technology on Parallel and Distributed Processing Laboratory Xiangtan University Xiangtan China The Artificial Intelligence Center PengCheng Laboratory shenzhen China Ajou University Suwon-si Republic of Korea

The following topics are dealt with: cloud computing; learning (artificial intelligence); resource allocation; security of data; mobile computing; neural nets; optimisation; parallel processing; scheduling; and storag... 详细信息

ISBN: (数字)9781728190747

ISBN: (纸本)9781728183824

关键词： Deep learning Schedules Multicore processing Logic gates Throughput Acceleration Field programmable gate arrays

来源：评论

学校读者我要写书评

暂无评论

Automatic generation and assessment of student assignments for parallel programming learning 10th

Automatic generation and assessment of student assignments f...

引用

10th International Symposium on parallel Architectures, Algorithms and Programming, PAAP 2019

作者： Luo, Zhenxiao Wang, Zelong Wu, Di Hei, Xiaojun Du, Yunfei School of Data and Computer Science Sun Yat-sen University Guangzhou China Guangdong Key Laboratory of Big Data Analysis and Processing Guangzhou510006 China School of Electronic Information and Communications Huazhong University of Science and Technology Wuhan China National Supercomputer Center in Guangzhou Guangzhou China

The course of parallel programming is becoming more and more important for the education of students majoring in computer science. However, it is not easy to learn parallel programming well due to its high theory and practice requirements. In this paper, we design and implement an automatic assignment generation and assessment system to help students learn parallel programming. The assignments can be generated according to user behaviors and thus able to guide students to learn parallel programming step by step. Besides, it can automatically generate an overall assessment of student assignments by using fuzzy string matching, which provides an approximate reference score of objective questions. Subjective questions can be assessed directly by comparing the answer to the reference answer. This system also provides a friendly user interface for students to complete online assignments and let teachers manage their question database. In our teaching practice, students can learn parallel programming more effectively with the help of such an assignment generation and assessment system. © Springer Nature Singapore Pte Ltd. 2020.

关键词： parallel programming

来源：评论

学校读者我要写书评

暂无评论

A multi-layered emulation platform to verify the effectiveness of consistency control approaches about message ordering

A multi-layered emulation platform to verify the effectivene...

引用

International Conference on Communication Software and Networks, ICCSN

作者： Hangjun Zhou Sha Fu Key Laboratory of Science and Technology for National Defence of Parallel and Distributed P rocessing School of Computer Science National University of Defense Technology Changsha Hunan China Department of Information Management Hunan University of Finance and Economics Changsha Hunan China

It is an important issue to preserve the consistent delivering order of messages at each site in distributed Virtual Environments(DVEs). Currently, the violations of message delivery order are inclined to happen in the large-scale and frequently interactive DVEs running on Wide Area Network(WAN). The inconsistent anomalies are characterized as unpredictable, unrepeatable and uncontrollable, which results in the severe problem for the development of the consistency control approaches. Meanwhile, it is also highly expensive to directly verify the proposed approaches on WAN. Therefore, in this paper, we design and establish an original multi-layered emulation platform in LAN, which can emulate the network latency of WAN and repeat the scenarios of violations. On the platform, the designers of consistency approaches don't have to spend efforts to realize the fundamental functions of the verification, but could pay more attention to the development and optimization of the control algorithms, which is very beneficial for the improvement of message ordering approaches.

关键词： Switches Wide area networks

来源：评论

学校读者我要写书评

暂无评论

RB2: Narrow the Gap between RDMA Abstraction and Performance via a Middle Layer

RB2: Narrow the Gap between RDMA Abstraction and Performance...

引用

IEEE Annual Joint Conference: INFOCOM, IEEE computer and Communications Societies

作者： Haifeng Sun Yixuan Tan Yongtong Wu Jiaqi Zhu Qun Huang Xin Yao Gong Zhang National Key Laboratory for Multimedia Information Processing School of Computer Science Peking University Huawei Theory Department

ISBN: (数字)9798350383508

ISBN: (纸本)9798350383515

Although the native RDMA interface allows for high throughput and low latency, its low-level abstraction raises significant programming challenges. Consequently, numerous systems encapsulate the RDMA interface into more user-friendly high-level abstractions such as Socket, MPI, and RPC. However, this ease of development often incurs considerable performance degradation. To address this trade-off, this paper introduces RB 2 , a high-performance RDMA-based distributed Ring Buffer (DRB). RB 2 serves as a middle layer that effectively conceals the low-level details of the RDMA interface while also facilitating extension to other high-level ***, it is non-trivial for DRBs to preserve the RDMA performance. We optimize the performance of RB 2 in three aspects. First, we perform micro-benchmarks to identify the pointer synchronization methods that are seemingly counter-intuitive but offer optimal performance improvements. Second, we propose an adaptive batching mechanism to alleviate the limitations of conventional fixed batching. Finally, we build an efficient memory subsystem using various optimization techniques. RB 2 outperforms SOTA designs by achieving 2.5 × to 7.5 × throughput while maintaining comparable tail latency for small messages.

关键词： Encapsulation Sockets Memory management Tail Programming Throughput Space exploration

来源：评论

学校读者我要写书评

暂无评论

Density-Attentive Head Detector for Crowd Counting

Density-Attentive Head Detector for Crowd Counting

引用

International Conference on Data science and Business Analytics (ICDSBA)

作者： Ziyu Zhao Xingyu Shen Long Lan Xiaoyao Yin Xiang Zhang Yonggang Che Zhigang Luo College of Computer National University of Defense Technology Changsha China Institute for Quantum Information & State Key Laboratory of High Performance Computing National University of Defense Technology Changsha China Science and Technology on Parallel and Distributed Processing National University of Defense Technology Changsha China

ISBN: (数字)9781728146447

ISBN: (纸本)9781728146454

In crowd counting, regression-based method shows better performance in extreme density scenes by introduces a density map. However, the regression-based method fails to locate the positions of each head, which significantly restricts its applications. Detection-based method counts each head with their accurate locations but works only in middle or low-level density scenes. In this paper, a joint learning method, named Density-attentive Head Detector (DAHD) is developed to overcome their respective shortcomings via a multi-task training procedure. Specifically, to guarantee the sensitivity of the detector to small or partially occluded heads, we carefully equip the detector with a density map which learned from regression module. Moreover, a novel Dilated Feature Pyramid Network (DFPN) is introduced to our method to enlarge the receptive field of convolutional kernel, bringing confirmative additional benefits to identify small heads. Experiments on the popular ShanghaiTech and Mall datasets confirm the improved performance of DAHD compared with the current detection-based approaches, and a comparable performance to regression-based approaches in term of counting.

关键词： Detectors Feature extraction Object detection Faces Kernel Training Task analysis

来源：评论

学校读者我要写书评

暂无评论

Eunomia: Scaling Concurrent Search Trees under Contention Using HTM

引用

ACM SIGPLAN Notices 2017年第8期52卷 385-399页

作者： Wang, Xin Zhang, Weihua Wang, Zhaoguo Wei, Ziyun Chen, Haibo Zhao, Wenyun Software School Fudan University China Shanghai Key Laboratory of Data Science Fudan University School of Computer Science Fudan University China Institute of Parallel and Distributed Systems Shanghai Jiao Tong University China Computer Science Department New York University United States

While hardware transactional memory (HTM) has recently been adopted to construct efficient concurrent search tree structures, such designs fail to deliver scalable performance under contention. In this paper, we first conduct a detailed analysis on an HTM-based concurrent B+Tree, which uncovers several reasons for excessive HTM aborts induced by both false and true conflicts under contention. Based on the analysis, we advocate Eunomia, a design pattern for search trees which contains several principles to reduce HTM aborts, including splitting HTM regions with version-based concurrency control to reduce HTM working sets, partitioned data layout to reduce false conflicts, proactively detecting and avoiding true conflicts, and adaptive concurrency control. To validate their effectiveness, we apply such designs to construct a scalable concurrent B+Tree using HTM. Evaluation using key-value store benchmarks on a 20-core HTM-capable multi-core machine shows that Eunomia leads to 5X-11X speedup under high contention, while incurring small overhead under low contention. © 2017 ACM.

关键词： Concurrency control

来源：评论

学校读者我要写书评

暂无评论

Completion time-a ware flow scheduling in heterogenous networks 1

引用

15th International Conference on Algorithms and Architectures for parallel processing, ICA3PP 2015

作者： He, Shiming Xie, Kun Zhang, Dafang Hunan Provincial Key Laboratory of Intelligent Processing of Big Data on Transportation School of Computer and Communication Engineering Hunan Province Engineering Research Center of Electric Transportation and Smart Distributed Network Changsha University of Science and Technology Changsha410114 China College of Computer Science and Electronics Engineering Hunan University Changsha410082 China Department of Electrical and Computer Engineering State University of New York at Stony Brook NY United States

ISBN: (数字)9783319271194

ISBN: (纸本)9783319271187

In China, the expressway isn’t free. When a vehicle exits, the exit toll station needs to calculate the toll according to the vehicle trajectory obtained by sending a trajectory query task to the trajectory center remotely. For the accurate of toll and keep the exit smoothly, transmitting the trajectory result reliably and in time is a key issue. There are three heterogenous networks in expressway, Fibernet, Telecommunications Network and 3G. We concurrently exploit three heterogenous networks by allocating the trajectory result into three sub-flows where each sub-flow is transmitted in one kind of networks. But existing multipath transfer solution(FMTCP) doesn’t consider the difference of completed time among all sub-flows and can’t make sure the completed time in time. Therefore, we formulate the sub-flows schedule as a integer programming problem, which is proved to be NP-hard problem, and propose a Completion time-aware Flow Scheduling scheme (CaFS) which calculates the size of sub-flows. For the reliably, it exploits generation-based network coding. For the transmission in time, CaFS allocates the size of sub-flow into suitable generation and scheduling the generations. The simulation results demonstrate that, comparing with the FMTCP, CaFS improves 31. 9% in completion time and 30% in task finished rate. © Springer International Publishing Switzerland 2015.

关键词： Network coding

来源：评论

学校读者我要写书评

暂无评论

A Unique Vertex Deleting Algorithm for Graph Isomorphism

A Unique Vertex Deleting Algorithm for Graph Isomorphism

引用

International Symposium on Image and Data Fusion (ISIDF)

作者： Baida Zhang Yuhua Tang Junjie Wu Linqi Huang National laboratory for Parallel and Distributed Processing School of Computer National University of Defense Technology Changsha Hunan P. R. China School of Resources and Safety Engineering Central South University Changsha P. R. China

Graph isomorphism problem has applications in many fields, such as chemistry, computer science, electronics, and network theory. But the exponential complexity of the algorithm makes the testing is time consuming. In this paper, a new algorithm named Unique Vertex Delete (UVD for short) is presented to speed up the process of isomorphic testing. The main idea of UVD algorithm is deleting vertex continuously to reduce the scale of the problem, and new field is added for each vertex to indicate the information of deleted neighbor vertices. Theoretical analysis and experiments show that the UVD algorithm consistently and significantly outperforms existing state-of-the-art approaches.

关键词： Algorithm design and analysis Testing Educational institutions Polynomials Software algorithms Complexity theory Partitioning algorithms

来源：评论

学校读者我要写书评

暂无评论

TRANSDUCTIVE NONNEGATIVE MATRIX FACTORIZATION FOR SEMI-SUPERVISED HIGH-PERFORMANCE SPEECH SEPARATION

TRANSDUCTIVE NONNEGATIVE MATRIX FACTORIZATION FOR SEMI-SUPER...

引用

IEEE International Conference on Acoustics, Speech and Signal processing

作者： Naiyang Guan Long Lan Dacheng Tao Zhigang Luo Xuejun Yang Science and Technology on Parallel and Distributed Processing Laboratory School of Computer Science National University of Defense Technology Changsha Hunan 410073 China Centre for Quantum Computation and Intelligent Systems FEIT University of Technology Sydney Sydney NSW 2007 Australia State Key Laboratory of High Performance Computing National University of Defense Technology Changsha Hunan 410073 China

ISBN: (纸本)9781479928941

Regarding the non-negativity property of the magnitude spectrogram of speech signals, nonnegative matrix factorization (NMF) has obtained promising performance for speech separation by independently learning a dictionary on the speech signals of each known speaker. However, traditional NM-F fails to represent the mixture signals accurately because the dictionaries for speakers are learned in the absence of mixture signals. In this paper, we propose a new transductive NMF algorithm (TNMF) to jointly learn a dictionary on both speech signals of each speaker and the mixture signals to be separated. Since TNMF learns a more descriptive dictionary by encoding the mixture signals than that learned by NMF, it significantly boosts the separation performance. Experiments results on a popular TIMIT dataset show that the proposed TNMF-based methods outperform traditional NMF-based methods for separating the monophonic mixtures of speech signals of known speakers.

关键词： Speech Dictionaries as Topic Loudspeakers NMF protocol signals mixtures factorization HIGH-PERFORMANCE

来源：评论

学校读者我要写书评

暂无评论

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

请选择保存的检索档案：

请选择收藏分类：

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

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

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：