检索结果-内蒙古大学图书馆

DLPlib： A Library for Deep Learning Processor

Journal of computer science & technology 2017年第2期32卷 286-296页

作者： Hui-Ying Lan Lin-Yang Wu Xiao Zhang Jin-Hua Tao Xun-Yu Chen Bing-Rui Wang Yu-Qing Wang Qi Guo Yun-Ji Chen State Key Laboratory of Computer Architecture Institute of Computing Technology Chinese Academy of Sciences Beijing 100190 China Microprocessor Research Center Institute of Computing Technology Chinese Academy of Sciences Beijing 100190 China University of Chinese Academy of Sciences Beijing 100049 China Department of Computer Science University of Science and Technology of China Hefei 230026 China

Recently, deep learning processors have become one of the most promising solutions of accelerating deep learning algorithms. Currently, the only method of programming the deep learning processors is through writing assembly instructions by bare hands, which costs a lot of programming efforts and causes very low efficiency. One solution is to integrate the deep learning processors as a new back-end into one prevalent high-level deep learning framework （e.g., TPU （tensor processing unit） is integrated into Tensorflow directly）. However, this will obstruct other frameworks to profit from the programming interface, The alternative approach is to design a framework-independent low-level library for deep learning processors （e.g., the deep learning library for GPU, cuDNN）. In this fashion, the library could be conveniently invoked in high-level programming frameworks and provides more generality. In order to allow more deep learning frameworks to gain benefits from this environment, we envision it as a low-level library which could be easily embedded into current high-level frameworks and provide high performance. Three major issues of designing such a library are discussed. The first one is the design of data structures. Data structures should be as few as possible while being able to support all possible operations. This will allow us to optimize the data structures easier without compromising the generality. The second one is the selection of operations, which should provide a rather wide range of operations to support various types of networks with high efficiency. The third is the design of the API, which should provide a flexible and user-friendly programming model and should be easy to be embedded into existing deep learning frameworks. Considering all the above issues, we propose DLPIib, a tensor-filter based library designed specific for deep learning processors. It contains two major data structures, tensor and filter, and a set of operators including basic neur

关键词： deep learning processor API library DLPlib

来源：评论

学校读者我要写书评

暂无评论

Big data-oriented energy prosumption service in smart community districts: a multi-case study perspective

引用

Energy Informatics 2019年第1期2卷 1-26页

作者： Anthony, Bokolo Petersen, Sobah Abbas Ahlers, Dirk Krogstie, John Livik, Klaus Department of Computer Science Norwegian University of Science and Technology NTNU TrondheimNO-7491 Norway Department of Architecture and Planning Norwegian University of Science and Technology Trondheim Norway Powel AS Klæbuveien 194 TrondheimNO-7037 Norway

The smart grid achieves bidirectional information and energy flow between energy consumer and utility grid, aiding energy users not only to utilize energy, but also to produce, sell, and share energy with other consumers or with the utility grid. This type of energy user is referred to as the "prosumer". Thus, prosumer management structures are important within energy market. However, prior studies on energy sustainability has paid little attention on prosumer involvement and management. Likewise, the continuous growth of cities has increased data processing complexity. Consequently, processing and analysis of historical, online, and real-time streaming data from energy sensors and metering devices has become a major issue in smart cities. Therefore, this research aims to present an architecture based on big data to improve energy prosumption in smart community districts by applying enterprise architecture approach grounded on The Open Group architecture Framework (TOGAF). Accordingly, qualitative methodology is adopted to collect data by employing case study by focus group interview from two energy companies in Norway to preliminarily validate the architecture. Findings from the case studies was demonstrated in ArchiMate modeling language to evaluate the applicability of the architecture. Moreover, findings from this study provides practical scenario that energy service providers can refer to in designing their own energy data platforms. Essentially, the architecture can be utilized as a guide to help municipalities and policy makers in creating approach for energy data analytics in smart community districts towards making decisions for future energy prosumption planning. © 2019, The Author(s).

关键词： Big data

来源：评论

学校读者我要写书评

暂无评论

Comparative Assessment Between Statistical and Soft Computing Methods for Accident Severity Classification

引用

Journal of The Institution of Engineers (India): Series A 2020年第1期101卷 27-40页

作者： Sarkar, Amrita Sarkar, Satyaki Department of Computer Science and Engineering Sarala Birla University Ranchi835103 India Department of Architecture Birla Institute of Technology Mesra Ranchi835215 India

Globally traffic scenario has experienced a vast change over the years as volumes of traffic on urban roads have increased by leaps and bounds. This increase has also marked an increase in the number of road accidents, both fatal and non-fatal. Accident prediction has a significant role in the improvement in traffic safety, and urban traffic management. Different statistical methods and various soft computing techniques are being used for developing accident prediction models. In this study, ensemble ranking method, comprising three filter-based ranking methods such as Information Gain, Gain Ratio, and Symmetric Uncertainty, has been employed to select significant features for prediction of accident severity. Through analysis, eight significant features have been selected for accident severity prediction. These are Annual Average Daily Traffic, Road Width, Percentage of Left-Turn Vehicles, Driver’s Age, Percentage of Right-Turn Vehicles, Types of Vehicles Involved, Spot Speed, and Cause of Accidents. Multiple logistic regression (MLGR) and artificial neural network (ANN) have been applied for accident severity prediction in urban roads. This analysis concludes that ANN model gives more accuracy and minimum errors compared with the MLGR method at the time of severity classification of a vehicular accident. © 2019, The Institution of Engineers (India).

关键词： Forecasting

来源：评论

学校读者我要写书评

暂无评论

Balanced Quantization： An Effective and Efficient Approach toQuantized Neural Networks

引用

Journal of computer science & technology 2017年第4期32卷 667-682页

作者： Shu-Chang Zhou Yu-Zhi Wang He Wen Qin-Yao He Yu-Heng Zou University of Chinese Academy of Sciences Beijing 100049 China State Key Laboratory of Computer Architecture Institute of Computing Technology Chinese Academy of Sciences Beijing 100190 China Megvii Inc. Beijing 100190 China Department of Electronic Engineering Tsinghua University Beijing 100084 China Department of Computer Science and Technology Tsinghua University Beijing 100084 China School of Electronics Engineering and Computer Science Peking University Beijing 100871 China

Quantized neural networks (QNNs), which use low bitwidth numbers for representing parameters and performing computations, have been proposed to reduce the computation complexity, storage size and memory usage. In QNNs, parameters and activations are uniformly quantized, such that the multiplications and additions can be accelerated by bitwise operations. However, distributions of parameters in neural networks are often imbalanced, such that the uniform quantization determined from extremal values may underutilize available bitwidth. In this paper, we propose a novel quantization method that can ensure the balance of distributions of quantized values. Our method first recursively partitions the parameters by percentiles into balanced bins, and then applies uniform quantization. We also introduce computationally cheaper approximations of percentiles to reduce the computation overhead introduced. Overall, our method improves the prediction accuracies of QNNs without introducing extra computation during inference, has negligible impact on training speed, and is applicable to both convolutional neural networks and recurrent neural networks. Experiments on standard datasets including ImageNet and Penn Treebank confirm the effectiveness of our method. On ImageNet, the top-5 error rate of our 4-bit quantized GoogLeNet model is 12.7%, which is superior to the state-of-the-arts of QNNs.

关键词： quantized neural network percentile histogram equalization uniform quantization

来源：评论

学校读者我要写书评

暂无评论

Boosting the Restoring Performance of Deduplication Data by Classifying Backup Metadata

引用

ACM/IMS Transactions on Data science 2021年第2期2卷 1–16页

作者： Yang, Ru Deng, Yuhui Zhou, Yi Huang, Ping Department of Computer Science Jinan University Guangzhou510632 China Department of Computer Science Jinan University China State Key Laboratory of Computer Architecture Institute of Computing Technology Chinese Academy of Sciences China TSYS School of Computer Science Columbus State University Doimukh Georgia College of Engineering Temple University United States

Restoring data is the main purpose of data backup in storage systems. The fragmentation issue, caused by physically scattering logically continuous data across a variety of disk locations, poses a negative impact on the restoring performance of a deduplication system. Rewriting algorithms are used to alleviate the fragmentation problem by improving the restoring speed of a deduplication system. However, rewriting methods give birth to a big sacrifice in terms of deduplication ratio, leading to a huge storage space waste. Furthermore, traditional backup approaches treat file metadata and chunk metadata as the same, which causes frequent on-disk metadata accesses. In this article, we start by analyzing storage characteristics of backup metadata. An intriguing finding shows that with 10 million files, the file metadata merely takes up approximately 340 MB. Motivated by this finding, we propose a Classified-Metadata based Restoring method (CMR) that classifies backup metadata into file metadata and chunk metadata. Because the file metadata merely takes up a meager amount of space, CMR maintains all file metadata in memory, whereas chunk metadata are aggressively prefetched to memory in a greedy manner. A deduplication system with CMR in place exhibits three salient features: (i) It avoids rewriting algorithms' additional overhead by reducing the number of disk reads in a restoring process, (ii) it increases the restoring throughput without sacrificing the deduplication ratio, and (iii) it thoroughly leverages the hardware resources to boost the restoring performance. To quantitatively evaluate the performance of CMR, we compare our CMR against two state-of-the-art approaches, namely, a history-aware rewriting method (HAR) and a context-based rewriting scheme (CAP). The experimental results show that compared to HAR and CAP, CMR reduces the restoring time by 27.2% and 29.3%, respectively. Moreover, the deduplication ratio is improved by 1.91% and 4.36%, respectively. © 2

关键词： Metadata

来源：评论

学校读者我要写书评

暂无评论

Reliability of complex production systems

引用

CIVIL ENGINEERING SYSTEMS 1996年第1期13卷 61-73页

作者： Kaplinski, O Milosz, M [a] Faculty of Architecture Civil andEnvironmental Engineering Poznah University of Technology Poznah Poland [b] Department of Computer Science Lublin Technical University Lublin Poland

This paper discusses problems of identifying and estimating the reliability of complex real production systems in the constructing industry. The influence of the system structure on its reliability has been considered. Types of elements and their connections in the system from the point of view of system reliability have been identified. The simulation method presented here allows to investigate the reliability of real systems, including the occurrence of inertia, variability of their structure and mutual dependence of reliability parameters of their elements. Two synthesis algorithms, (considering phase inertial in the estimation of system reliability), have been proposed and compared. On the basis of a machine system (loader-conveyor belt-material receiver), the utilisation of inertia to improve reliability characteristics of the system has been proved to be more effective than adding reserve elements to the system.

关键词： reliability system structure decomposition and synthesis simulation construction operations

来源：评论

学校读者我要写书评

暂无评论

Register-based implementation of the sparse general matrix-matrix multiplication on GPUs

引用

ACM SIGPLAN Notices 2018年第1期53卷 407-408页

作者： Liu, Junhong He, Xin Liu, Weifeng Tan, Guangming State Key Laboratory of Computer Architecture Institute of Computing Technology Chinese Academy of Sciences China University of Chinese Academy of Sciences China Department of Computer Science Norwegian University of Science and Technology Norway

General sparse matrix-matrix multiplication (SpGEMM) is an essential building block in a number of applications. In our work, we fully utilize GPU registers and shared memory to implement an efficient and load balance... 详细信息

ISBN: (纸本)9781450349116

关键词： Matrix algebra

来源：评论

学校读者我要写书评

暂无评论

Sampling Methods for Efficient Training of Graph Convolutional Networks:A Survey

引用

IEEE/CAA Journal of Automatica Sinica 2022年第2期9卷 205-234页

作者： Xin Liu Mingyu Yan Lei Deng Guoqi Li Xiaochun Ye Dongrui Fan State Key Laboratory of Computer Architecture Institute of Computing TechnologyChinese Academy of SciencesBeijing 100086 School of Computer Science and Technology University of Chinese Academy of SciencesBeijing 100049 IEEE Department of Precision Instrument Center for Brain Inspired Computing ResearchTsinghua UniversityBeijing 100084China

Graph convolutional networks(GCNs)have received significant attention from various research fields due to the excellent performance in learning graph *** GCN performs well compared with other methods,it still faces *** a GCN model for large-scale graphs in a conventional way requires high computation and storage ***,motivated by an urgent need in terms of efficiency and scalability in training GCN,sampling methods have been proposed and achieved a significant *** this paper,we categorize sampling methods based on the sampling mechanisms and provide a comprehensive survey of sampling methods for efficient training of *** highlight the characteristics and differences of sampling methods,we present a detailed comparison within each category and further give an overall comparative analysis for the sampling methods in all ***,we discuss some challenges and future research directions of the sampling methods.

关键词： Efficient training graph convolutional networks(GCNs) graph neural networks(GNNs) sampling method

来源：评论

学校读者我要写书评

暂无评论

Local image phase, energy and orientation extraction using FPGAs

Local image phase, energy and orientation extraction using F...

引用

作者： Díaz, J. Ros, E. Mota, S. Carrillo, R. Department of Computer Architecture and Technology University of Granada Granada Spain Department of Computer Science and Numerical Analysis University of Cordoba Cordoba Spain

Local features such as phase, energy and orientation have been widely used in the literature for multiple computer vision applications such as edge detection, texture analysis, motion estimation, depth computation, etc. They can be computed using wavelet image decomposing based on Quadrature filters tuned at different scales and orientations. Although this represents a very powerful image analysis technique, its feasibility for real-time domain applications is still beyond current computer capabilities due to the high demand of computational resources. In this article, we present an architecture developed for real-time processing of these features using reconfigurable hardware. Although, to the best of our knowledge, previous implementations of Gabor or steerable filters on chip already exist, none of them provides energy, phase, and orientation information at the same time. We show how to implement quadrature filters based on Gaussian derivatives. We develop a hardware friendly technique to interpolate between a multi-oriented filter bank to compute features at their main orientation which has not been used before. We finally describe the system resource requirements, performance and some system illustrative results. © 2008 Taylor & Francis.

关键词： Fixed point arithmetic

来源：评论

学校读者我要写书评

暂无评论

Fuzzy partitions for texture modelling adapted to the subjectivity of human perception

Fuzzy partitions for texture modelling adapted to the subjec...

引用

2016 IEEE International Conference on Fuzzy Systems, FUZZ-IEEE 2016

作者： Pedro, Manuel Martínez-Jiḿenez Jeśus, Chamorro-Martínez Jośe, Manuel Soto-Hidalgo Department of Computer Science and Artificial Intelligence University of Granada Spain Department of Computer Architecture Electronics and Electronic Technology University of Ćordoba Spain

ISBN: (纸本)9781509006250

Visual textures in images are usually described by humans using linguistic terms related to their perceptual properties, like "very coarse", "low directional", or "high contrasted". However, the presence of these properties in images is very difficult to characterize due to their imprecision, and, moreover, because their perception may change depending on the user. In this paper, we propose an adaptive fuzzy approach for texture modelling, that allows to describe it in natural language, and takes into account the subjectivity of the human perception. For this purpose, a methodology in two phases has been proposed. First, non-adaptive fuzzy partitions, that represent the average human perception about the presence of the texture properties, are obtained. For this modelling, a perception-based fuzzy technique have been used. In a second phase, the obtained fuzzy sets are adapted in order to model the particular perception that a user may have. For this purpose, the membership functions are automatically transformed on the basic of the information given by the user to represent his perception. © 2016 IEEE.

关键词： Membership functions

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：