检索结果-内蒙古大学图书馆

Implementation of the parallel mean shift-based imagesegmentation algorithm on a GPU cluster

学校读者我要写书评

暂无评论

International Journal of Digital Earth 2019年第3期12卷 328-353页

作者： Fang Huang Yinjie Chen Li Li Ji Zhou Jian Tao Xicheng Tan Guangsong Fana School of Resources&Environment University of Electronic Science and Technology of ChinaChengduPeople’s Republic of China Institute of Remote Sensing Big Data Big Data Research CenterUniversity of Electronic Science and Technology of ChinaChengduPeople’s Republic of China School of Construction&Management Engineering Xihua UniversityChengduPeople’s Republic of China Texas A&M Engineering Experiment Station and High Performance Research Computing Texas A&M UniversityCollege StationTXUSA International School of Software Wuhan UniversityWuhanPeople’s Republic of China

The mean shift image segmentation algorithm is very computationintensive. To address the need to deal with a large number of remotesensing (RS) image segmentations in real-world applications, this studyhas investigated the parallelization of the mean shift algorithm on asingle graphics processing unit (GPU) and a task-scheduling methodwith message passing interface (MPI)+OpenCL programming model on aGPU cluster platform. This paper presents the test results of the parallelmean shift image segmentation algorithm on Shelob, a GPU clusterplatform at Louisiana State University, with different datasets andparameters. The experimental results show that the proposed parallelalgorithm can achieve good speedups with different configurations andRS data and can provide an effective solution for RS image processingon a GPU cluster.

关键词： Mean shift algorithm GPU cluster task scheduling MPI OpenCL

Density Fluctuations in the Intracluster Medium: An Attempt to Constrain Viscosity with Cosmological Simulations

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Marin-Gilabert, Tirso Steinwandel, Ulrich P. Valentini, Milena Vallés-Pérez, David Dolag, Klaus Universitäts-Sternwarte Fakultät für Physik Ludwig-Maximilians-Universität München Scheinerstr.1 München81679 Germany Center for Computational Astrophysics Flatiron Institute 162 5th Avenue New YorkNY10010 United States Astronomy Unit Department of Physics University of Trieste via Tiepolo 11 TriesteI-34131 Italy INAF - Osservatorio Astronomico di Trieste via Tiepolo 11 TriesteI-34131 Italy INFN Instituto Nazionale di Fisica Nucleare Via Valerio 2 TriesteI-34127 Italy ICSC - Italian Research Center on High Performance Computing Big Data and Quantum Computing via Magnanelli 2 Casalecchio di Reno40033 Italy Departament d'Astronomia i Astrofísica Universitat de València C/Doctor Moliner 50 València BurjassotE-46100 Spain Max-Planck-Institut für Astrophysik Karl-Schwarzschild-Straße 1 Garching85741 Germany

The impact of viscosity in the Intracluster Medium (ICM) is still an open question in astrophysics. To address this problem, we have run a set of cosmological simulations of three galaxy clusters with a mass larger than MVir > 1015M☉ at z = 0 using the SPMHD-code OpenGadget3. We aim to quantify the influence of viscosity and constrain its value in the ICM. Our results show significant morphological differences at small scales, temperature variations, and density fluctuations induced by viscosity. We observe a suppression of instabilities at small scales, resulting in a more filamentary structure and a larger amount of small structures due to the lack of mixing with the medium. The conversion of kinetic to internal energy leads to an increase of the virial temperature of the cluster of ∼5% - 10%, while the denser regions remain cold. The amplitude of density and velocity fluctuations are found to increase with viscosity. However, comparison with observational data indicates that the simulations, regardless of the viscosity, match the observed slope of the amplitude of density fluctuations, challenging the direct constraint of viscosity solely through density fluctuations. Furthermore, the ratio of density to velocity fluctuations remains close to 1 regardless of the amount of viscosity, in agreement with the theoretical expectations. Our results show for the first time in a cosmological simulation of a galaxy cluster the effect of viscosity in the ICM, a study that is currently missing in the literature. Copyright © 2024, The Authors. All rights reserved.

关键词： Cosmology

An implementation of nDGP gravity in Pinocchio

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Song, Yanling Hu, Bin Ruan, Cheng-Zong Moretti, Chiara Monaco, Pierluigi Center for Gravitation and Cosmology College of Physical Science and Technology Yangzhou University Yangzhou225009 China Institute for Frontier in Astronomy and Astrophysics Beijing Normal University Beijing102206 China Department of Astronomy Beijing Normal University Beijing100875 China Institute of Theoretical Astrophysics University of Oslo Oslo0315 Norway SISSA - International School for Advanced Studies Via Bonomea 265 Trieste34136 Italy Centro Nazionale "High Performance Computer Big Data and Quantum Computing" Italy INAF Osservatorio Astronomico di Trieste Via Tiepolo 11 TriesteI-34143 Italy IFPU Institute for Fundamental Physics of the Universe Via Beirut 2 Trieste34014 Italy Dipartimento di Fisica Universitá di Trieste Sezione di Astronomia via Tiepolo 11 TriesteI-34143 Italy INFN Sezione di Trieste Italy

In this paper we investigate dark matter structure formation in the normal branch of the Dvali-Gabadadze-Porrati (nDGP) model using the PINOCCHIO algorithm. We first present 2nd order Lagrangian perturbation theory for the nDGP model, which shows that the 1st- and 2nd-order growth functions in nDGP are larger than those in ΛCDM. We then examine the dynamics of ellipsoidal collapse in nDGP, which is accelerated compared to ΛCDM due to enhanced gravitational interactions. Running the nDGP-PINOCCHIO code with a box size of 512 Mpc h−1 and 10243 particles, we analyze the statistical properties of the output halo catalogs, including the halo power spectrum and halo mass function. The calibrated PINOCCHIO halo power spectrum agrees with N-body simulations within 5% in the comoving wavenumber range k −1) at redshift z = 0. The agreement is extended to smaller scales for higher redshifts. For the cumulative halo mass function, the agreement between N-body and PINOCCHIO is also within the simulation scatter. Copyright © 2023, The Authors. All rights reserved.

关键词： Power spectrum

DSRGAN: Explicitly learning disentangled representation of underlying structure and rendering for image generation without tuple supervision

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Hao, Guang-Yuan Yu, Hong-Xing Zheng, Wei-Shi School of Data and Computer Science Sun Yat-sen University Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Collaborative Innovation Center of High Performance Computing Nudt Guangdong Key Laboratory of Big Data Analysis and Processing Guangzhou

We focus on explicitly learning disentangled representation for natural image generation, where the underlying spatial structure and the rendering on the structure can be independently controlled respectively, yet using no tuple supervision. The setting is significant since tuple supervision is costly and sometimes even unavailable. However, the task is highly unconstrained and thus ill-posed. To address this problem, we propose to introduce an auxiliary domain which shares a common underlying-structure space with the target domain, and we make a partially shared latent space assumption. The key idea is to encourage the partially shared latent variable to represent the similar underlying spatial structures in both domains, while the two domain-specific latent variables will be unavoidably arranged to present renderings of two domains respectively. This is achieved by designing two parallel generative networks with a common Progressive Rendering Architecture (PRA), which constrains both generative networks' behaviors to model shared underlying structure and to model spatially dependent relation between rendering and underlying structure. Thus, we propose DSRGAN (GANs for Disentangling Underlying Structure and Rendering) to instantiate our method. We also propose a quantitative criterion (the Normalized Disentanglability) to quantify disentanglability. Comparison to the state-of-The-Art methods shows that DSRGAN can significantly outperform them in disentanglability. Copyright © 2019, The Authors. All rights reserved.

关键词： Rendering (computer graphics)

IndexIt: Enhancing data locating services for parallel file systems 21

学校读者我要写书评

暂无评论

IndexIt: Enhancing data locating services for parallel file ...

21st IEEE International Conference on high performance computing and Communications, 17th IEEE International Conference on Smart City and 5th IEEE International Conference on data Science and Systems, HPCC/SmartCity/DSS 2019

作者： Cheng, Peng Wang, Yong Lu, Yutong Du, Yunfei Chen, Zhiguang College of Computer National University of Defense Technology Changsha China State Key Laboratory of High Performance Computing Changsha China National Supercomputer Center in Guangzhou School of Data and Computer Science Sun Yat-Sen University Guangzhou China

ISBN: (纸本)9781728120584

While the ability to access a small fraction of data records from a large volume of scientific datasets is vital to accelerate scientific discovery, existing parallel file systems face serious challenges in managing scientific big data since data services have traditionally been decoupled from file systems. In this paper, we present IndexIt, an in-situ index and query middleware that aims to enhance record locating services for parallel file systems. When applications are writing data to parallel file systems, IndexIt allows users to index data while they are still in memory and bypass the performance bottleneck between memory and disks. By applying lightweight Bitmap-Range index and organizing the index as key-value pairs, IndexIt accelerates index building efficiently. Moreover, we propose a two-level query processing framework to process query requests. We build a prototype and evaluate its performance with real scientific datasets. Compared with existing data management tools, the index building time can be reduced dramatically. The proposed two-level query processing achieves up to 3 orders of magnitude speed-up than scanning the entire dataset and shows up to 1.35x speed-up than an existing tool. © 2019 IEEE.

关键词： Information management

Gradient Amplification: An efficient way to train deep neural networks

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Basodi, Sunitha Ji, Chunyan Zhang, Haiping Pan, Yi Department of Computer Science Georgia State University AtlantaGA30302 United States Center for High Performance Computing Joint Engineering Research Center for Health Big Data Intelligent Analysis Technology Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences ShenzhenGuangdong518 055 China

Improving performance of deep learning models and reducing their training times are ongoing challenges in deep neural networks. There are several approaches proposed to address these challenges one of which is to increase the depth of the neural networks. Such deeper networks not only increase training times, but also suffer from vanishing gradients problem while training. In this work, we propose gradient amplification approach for training deep learning models to prevent vanishing gradients and also develop a training strategy to enable or disable gradient amplification method across several epochs with different learning rates. We perform experiments on VGG-19 and resnet (Resnet-18 and Resnet-34) models, and study the impact of amplification parameters on these models in detail. Our proposed approach improves performance of these deep learning models even at higher learning rates, thereby allowing these models to achieve higher performance with reduced training time. Copyright © 2020, The Authors. All rights reserved.

关键词： Backpropagation

Correction to “Privacy-Preserving Blockchain-Based Federated Learning for IoT Devices” [Feb 20 1817-1829]

学校读者我要写书评

暂无评论

IEEE Internet of Things Journal 2022年第1期10卷 973-973页

作者： Yang Zhao Jun Zhao Linshan Jiang Rui Tan Dusit Niyato Zengxiang Li Lingjuan Lyu Yingbo Liu School of Computer Science and Engineering Nanyang Technological University Jurong West Singapore Institute of High Performance Computing A*STAR Fusionopolis Singapore Department of Computer Science National University of Singapore Queenstown Singapore Big Data Research Institute of Yunnan Economy and Society Yunnan University of Finance and Economics Kunming China

In [1] , on page 1824, Fig. 3 should be as follows:

关键词： Federated learning Internet of Things Bockchains Differential learning Crowdsourcing

An auto code generator for stencil on SW26010 21

学校读者我要写书评

暂无评论

An auto code generator for stencil on SW26010

作者： Zhu, Xiaomin Zeng, Yunhui Wei, Yanjie Feng, Shengzhong Liu, Weiguo Balaji, Pavan Joint Engineering Research Center Health Big Data Intelligent Analysis Technology Center for High Performance Computing Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences Shenzhen China Shandong Computer Science Center National SuperComputer Center in Jinan Jinan China Sclool of Software Shandong University Jinan China Argonne National Laboratory United States

ISBN: (纸本)9781728120584

Stencil is a basic building block widely used in many HPC areas and applications. It generally dominates the time cost and is critical to the overall performance. Given that heterogeneous many-core is frequently adopted to build super computers, porting and optimizing stencil codes on modern accelerator-based architectures is important. Porting is non-trivial, and optimization is more difficult as it requires better understanding of the underlying architecture. As a result, auto tuning targeting on accelerators such as GPU becomes a hot research topic. In this paper, we focus on tuning stencil automatically on SW26010, which is a heterogeneous many-core CPU equipped in Sunway TaihuLight SuperComputer, and mainly work on 2D stencil variants code in ROMS (Regional Ocean Modeling System). The code generator can generate codes running on slave cores for 50+ different FORTRAN stencil-alike loops in ROMS in seconds, and optimization methods are integrated into the generator as well. The performance is better than the directive based method (OpenACC) and the hand-drafted code by a junior programmer, and it is comparable compared with the code written by a senior programmer with several years of programming experience. © 2019 IEEE.

关键词： Supercomputers

Cross-lingual machine reading comprehension with language branch knowledge distillation

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Liu, Junhao Shou, Linjun Pei, Jian Gong, Ming Yang, Min Jiang, Daxin Shenzhen Key Laboratory for High Performance Data Mining Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences China University of Chinese Academy of Sciences China Microsoft STCA NLP Group Beijing China School of Computing Science Simon Fraser University Canada

Cross-lingual Machine Reading Comprehension (CLMRC) remains a challenging problem due to the lack of large-scale annotated datasets in low-source languages, such as Arabic, Hindi, and Vietnamese. Many previous approaches use translation data by translating from a rich-source language, such as English, to low-source languages as auxiliary supervision. However, how to effectively leverage translation data and reduce the impact of noise introduced by translation remains onerous. In this paper, we tackle this challenge and enhance the cross-lingual transferring performance by a novel augmentation approach named Language Branch Machine Reading Comprehension (LBMRC). A language branch is a group of passages in one single language paired with questions in all target languages. We train multiple machine reading comprehension (MRC) models proficient in individual language based on LBMRC. Then, we devise a multilingual distillation approach to amalgamate knowledge from multiple language branch models to a single model for all target languages. Combining the LBMRC and multilingual distillation can be more robust to the data noises, therefore, improving the model’s cross-lingual ability. Meanwhile, the produced single multilingual model is applicable to all target languages, which saves the cost of training, inference, and maintenance for multiple models. Extensive experiments on two CLMRC benchmarks clearly show the effectiveness of our proposed method. Copyright © 2020, The Authors. All rights reserved.

关键词： Large dataset