检索结果-内蒙古大学图书馆

33rd International Conference on Machine Learning, ICML 2016

作者： Yang, Tianbao Zhang, Lijun Jin, Rong Yi, Jinfeng Department of Computer Science University of Iowa Iowa CityIA52242 United States National Key Laboratory for Novel Software Technology Nanjing University Nanjing China Institute of Data Science and Technologies Alibaba Group Seattle United States IBM Thomas J. Watson Research Center Yorktown HeightsNY10598 United States

ISBN: (纸本)9781510829008

This work focuses on dynamic regret of online convex optimization that compares the performance of online learning to a clairvoyant who knows the sequence of loss functions in advance and hence selects the minimizer of the loss function at each step. By assuming that the clairvoyant moves slowly (i.e., the minimizers change slowly), we present several improved variationbased upper bounds of the dynamic regret under the true and noisy gradient feedback, which are optimal in light of the presented lower bounds. The key to our analysis is to explore a regularity metric that measures the temporal changes in the clairvoyant's minimizers, to which we refer as path variation. Firstly, we present a general lower bound in terms of the path variation, and then show that under full information or gradient feedback we are able to achieve an optimal dynamic regret. Secondly, we present a lower bound with noisy gradient feedback and then show that we can achieve optimal dynamic regrets under a stochastic gradient feedback and two-point bandit feedback. Moreover, for a sequence of smooth loss functions that admit a small variation in the gradients, our dynamic regret under the two-point bandit feedback matches what is achieved with full information.

关键词： Stochastic systems

来源：评论

学校读者我要写书评

暂无评论

A sample weighted sparse representation approach to image classification

引用

ICIC Express Letters, Part B: Applications 2016年第11期7卷 2283-2290页

作者： Wang, Yongxin Zhang, Huaxiang Department of Computer Science Shandong Normal University China Shandong Provincial Key Laboratory for Novel Distributed Computer Software Technology No. 88 Wenhua East Road Lixia District Jinan250014 China

Sparse representation for classification (SRC) has achieved a big success for face recognition. It utilizes a sparsely linear combination of the training samples to construct a test sample, and classifies the test sample based on the reconstruction error associated with each class. Since SRC weights each training sample equally important, which may not hold for real applications, we propose a novel sample weighted sparse representation classification approach (SWSRC) by weighting each training sample differently. We first employ the representation ability of each training sample to construct a weight matrix and then solve a weighted l1minimization problem to obtain the sparse reconstruction coefficients. Experimental results on AR, YaleB and USPS image datasets demonstrate its effectiveness. © 2016 ICIC International.

关键词： Face recognition

来源：评论

学校读者我要写书评

暂无评论

A new data replication scheme for PVFS2 15th

A new data replication scheme for PVFS2

引用

15th International Conference on Algorithms and Architectures for Parallel Processing, ICA3PP 2015

作者： Bao, Nianyuan Tang, Jie Zhang, Xiaoyu Wu, Gangshan State Key Laboratory for Novel Software Technology Department of Computer Science and Technology Nanjing University Nanjing210046 China

ISBN: (纸本)9783319271361

PVFS is one of the most popular distributed file systems with parallelism, which is still widely used today. Now PVFS is in its version 2, called PVFS2. PVFS2 has a leading performance on I/O operations, but the reliability and stability are not as good. One of the reasons is the lack of data replication. This paper presents a new data replication scheme in PVFS2. In our approach, the backup operation is done on the servers, therefore the user experience is not affected while creating copies of files. In addition, we optimized the read operation of PVFS2. With copies, we can choose the servers to read from, so we can maintain parallelism of read operation under complex conditions such as a server is down or the load of some servers are obviously higher than others. Experimental results verify the effectiveness and efficiency of our method. © Springer International Publishing Switzerland 2015.

关键词： File organization

来源：评论

学校读者我要写书评

暂无评论

Pre-stack kirchhoff time migration on hadoop and spark 15th

Pre-stack kirchhoff time migration on hadoop and spark

引用

15th International Conference on Algorithms and Architectures for Parallel Processing, ICA3PP 2015

作者： Yang, Chen Tang, Jie Gao, Heng Wu, Gangshan State Key Laboratory for Novel Software Technology Department of Computer Science and Technology Nanjing University Nanjing210046 China

ISBN: (纸本)9783319271361

Pre-stack Kirchhoff time migration (PKTM) is one of the most widely used migration algorithms in seismic imaging area. How- ever, PKTM takes considerable time due to its high computational cost, which greatly affects the working efficiency of oil industry. Due to its high fault tolerance and scalability, Hadoop has become the most popular platform for big data processing. To overcome the shortcoming too much network traffic and disk I/O in Hadoop, there shows up a new distributed framework—Spark. However the behaviour and performance of those two systems when applied to high performance computing are still under investigation. In this paper, we proposed two parallel algorithms of the plrestack Kirchhoff time migration based on Hadoop and Sark respectively. Experiments are carried out to compare the performances of them. The results show that both of implementations are efficient and scalable and our PKTM on Spark exhibits better performance than the one on Hadoop. © Springer International Publishing Switzerland 2015.

关键词： MapReduce

来源：评论

学校读者我要写书评

暂无评论

A dynamic extension and data migration method based on PVFS 1

引用

15th International Conference on Algorithms and Architectures for Parallel Processing, ICA3PP 2015

作者： Zhang, Xiaoyu Tang, Jie Gao, Heng Wu, Gangshan State Key Laboratory for Novel Software Technology Department of Computer Science and Technology Nanjing University Nanjing210046 China

ISBN: (数字)9783319271224

ISBN: (纸本)9783319271217

With the development of the big data, The traditional file system can no longer meet the demand of High Performance Computing and Big Data. Parallel file systems are getting more and more popular in High Performance Computing. As a typical parallel file system, PVFS has been widely used in big data computing area in recent years. However with the increasing of computing scale, there exist the needs to dynamic extend data nodes, which PVFS does not support at present. This paper put forward a dynamic data node extension method as well as the subsequent data migration algorithm based on PVFS. The algorithm first adds a new data node automatically and transparently. After that, the algorithm finds out the most loaded data node in the original file system using a new load evaluation method and transfer the data into the newly added data node to mitigate the imbalance of the system. The experimental results show that our dynamic data node extension method could improve the performance of PVFS and reduce the probability of hot point effectively. © Springer International Publishing Switzerland 2015.

关键词： Big data

来源：评论

学校读者我要写书评

暂无评论

Compile-Time Automatic Synchronization Insertion and Redundant Synchronization Elimination for GPU Kernels

Compile-Time Automatic Synchronization Insertion and Redunda...

引用

International Conference on Parallel and Distributed Systems (ICPADS)

作者： Lifeng Liu Meilin Liu Chong-Jun Wang Jun Wang Department of Computer Science and Engineering Wright State University Dayton OH USA National Key Laboratory for Novel Software Technology Nanjing University Nanjing China Department of Electrical Engineering and Computer Science University of Central Florida Orlando FL USA

ISBN: (纸本)9781509053827

In most of the GPU kernel programs, the synchronization statements are inserted manually by the programmers, which is very labor intensive, and error-prone. In this paper, we propose a synchronization optimization framework to automatically insert synchronization statements into the GPU kernels at compile time, while eliminating the redundant synchronization statements. We have shown that our framework can not only insert the synchronizations correctly, but also eliminate the redundant synchronizations, which outperforms the existing compiler frameworks that introduce redundant synchronizations using the most conservative strategy. Taking the GPU kernels as the input, our framework leverages data dependence analysis to insert synchronizations. We extend CETUS, a source-to-source compiler framework, to implement our synchronization optimization framework. Experimental results show that our proposed framework achieved 100% correctness by combining extensive evaluation and manual comparison. In addition, the number of synchronization statements in GPU kernels is reduced by 32.5%, and the number of synchronization statements executed is reduced by 28.2% on average by our synchronization optimization framework compared to the original GPU kernels.

关键词： Synchronization Graphics processing units Kernel Optimization computer architecture Indexes Algorithm design and analysis

来源：评论

学校读者我要写书评

暂无评论

An ant colony optimization algorithm for learning brain effective connectivity network from fMRI data

An ant colony optimization algorithm for learning brain effe...

引用

IEEE International Conference on Bioinformatics and Biomedicine (BIBM)

作者： Jinduo Liu Junzhong Ji Aidong Zhang Peipeng Liang Beijing Municipal Key Laboratory of Multimedia and Intelligent Software the College of Computer Science and Technology Beijing University of Technology Beijing China Department of Computer Science and Engineering University at Buffalo The State University of New York Buffalo America Beijing Key Lab of MRI and Brain Informatics Department of Radiology Capital Medical University Beijing China

ISBN: (纸本)9781509016129

Identifying brain effective connectivity networks from functional magnetic resonance imaging (fMRI) data is an important advanced subject in neuroinformatics in recent years, where the learning method based on bayesian networks (BN) has become a new hot topic in the field. This paper proposes a new method to learn the brain effective connectivity network structure by combining ant colony optimization (ACO) with BN method, named as ACOEC. In the proposed algorithm, a brain effective connectivity network is first mapped onto an ant, and then the ant colony optimization by simulating real ants looking for food is employed to construct network structures and finally an ant with the highest score is obtained as the optimal solution. The experimental results on simulated and real fMRI data sets show that the new method can not only accurately identify the connections and directions of the brain networks, but also quantitatively describe the connection strength of the brain networks, which has a good clinical application prospects.

关键词： Ant colony optimization Bayes methods Measurement Mathematical model Brain modeling Optimization Electronic mail

来源：评论

学校读者我要写书评

暂无评论

Fast RFID Polling Protocols

Fast RFID Polling Protocols

引用

International Conference on Parallel Processing (ICPP)

作者： Jia Liu Bin Xiao Xuan Liu Lijun Chen State Key Laboratory for Novel software Technology Nanjing University China Department of Computing The Hong Kong Polytechnic University China College - of Computer Science and Electronic Engineering Hunan University China

ISBN: (纸本)9781509028245

Polling is a widely used anti-collision protocol that interrogates RFID tags in a request-response way. In conventional polling, the reader needs to broadcast 96-bit tag IDs to separate each tag from others, leading to long interrogation delay. This paper takes the first step to design fast polling protocols by shortening the polling vector. We first propose an efficient Hash Polling Protocol (HPP) that uses hash indices rather than tag IDs as the polling vector to query each tag. The length of the polling vector is dropped from 96 bits to no more than log(n) bits (n is the number of tags). We then enhance HPP (EHPP) to make it not only more efficient but also more steady with respect to the number of tags. To avoid redundant transmissions in both HPP and EHPP, we finally propose a Tree-based Polling Protocol (TPP) that reserves the invariant portion of the polling vector while updates only the discrepancy by constructing and broadcasting a polling tree. Theoretical analysis shows that the average length of the polling vector in TPP levels off at only 3.44, 28 times less than 96-bit tag IDs. We also apply our protocols to collect tag information and simulation results demonstrate that our best protocol TPP outperforms the state-of-the-art information collection protocol.

关键词： Protocols Indexes RFID tags Broadcasting Wireless communication Delays

来源：评论

学校读者我要写书评

暂无评论

Throughput Optimization in Cognitive Radio Networks Ensembling Physical Layer Measurement

引用

Journal of computer science & technology 2015年第6期30卷 1290-1305页

作者：赵彦超吴杰李文中陆桑璐 State Key Laboratory for Novel Software Technology Nanjing University Nanjing 210023 China Department of Computer and Information Sciences Temple University Philadelphia PA 19122 U.S.A.

Wireless networks are developed under the fashion of wider spectrum utilization （e.g., cognitive radio） and multi-hop communication （e.g., wireless mesh networks）. In these paradigms, how to effectively allocate the spectrum to different transmission links with minimized mutual interference becomes the key concern. In this paper, we study the throughput optimization via spectrum allocation in cognitive radio networks （CRNs）. The previous studies incorporate either the conflict graph or SINR model to characterize the interference relationship. However, the former model neglects the accumulative interference effect and leads to unwanted interference and sub-optimal results, while the work based on the latter model neglects its heavy reliance on the accuracy of estimated RSS （receiving signal strength） among all potential links. Both are inadequate to characterize the complex relationship between interference and throughput. To this end, by considering the feature of CRs, like spectrum diversity and non-continuous OFDM, we propose a measurement-assisted SINR-based cross-layer throughput optimization solution. Our work concerns features in different layers： in the physical layer, we present an efficient RSS estimation algorithm to improve the accuracy of the SINR model; in the upper layer, a flow level SINR-based throughput optimization problem for WMNs is modelled as a mixed integer non-linear programming problem which is proved to be NP-hard. To solve this problem, a centralized （1 -ε）-optimal algorithm and an efficient distributed algorithm are provided. To evaluate the algorithm performance, the real-world traces are used to illustrate the effectiveness of our scheme.

关键词： cognitive radio network wireless mesh network throughput optimization centralized algorithm distributedalgorithm spectrum allocation

来源：评论

学校读者我要写书评

暂无评论

Smart Grid Data Analysis and Prediction Modeling

Smart Grid Data Analysis and Prediction Modeling

引用

2016 IEEE International Conference on Progress in Informatics and Computing

作者： Hang Yang Ping Li Anum Masood Yuning Xiao Bin Sheng Qichen Yu Information Department China Southern Power Grid Department of Mathematics and Information Technology The Education University of Hong Kong Department of Computer Science and Engineering Shanghai Jiao Tong University Department of Computer Science COMSATS Institute of Information Technology The State Key Laboratory of Computer Science Institute of Software Chinese Academy of Sciences Department of electrical engineering and automation Shanghai Jiao Tong University

With the rapiddevelopment of power grid, prediction ofelectric quantity changes has become increasingly important. High-performance power grid systems can improve economic effectiveness and operational efficiency through accurate prediction. This paper proposes a prediction model based on temperature, humidity,time, and the number of people. On account of the standards of support vector machine(SVM) and the HBase platform,we have implemented a forecasting model and designed simulative experiments. The experimental results show that time and variation in the number of people has a remarkable influence on prediction, while temperature and humidityhardly have any effects.

关键词： power prediction data mining HBase data storage machine learning SVM

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：