检索结果-内蒙古大学图书馆

ACM/IEEE International Conference for High Performance Computing, Networking, Storage, and Analysis

作者： Evangelos Georganas Sasikanth Avancha Kunal Banerjee Dhiraj Kalamkar Greg Henry Hans Pabst Alexander Heinecke Intel Corporation

ISBN: (纸本)9781538683859

Convolution layers are prevalent in many classes of deep neural networks, including convolutional Neural Networks (CNNs) which provide state-of-the-art results for tasks like image recognition, neural machine translation and speech recognition. The computationally expensive nature of a convolution operation has led to the proliferation of implementations including matrix-matrix multiplication formulation, and direct convolution primarily targeting GPUs. In this paper, we introduce direct convolution kernels for ×86 architectures, in particular for Xeon and Xeon Phi systems, which are implemented via a dynamic compilation approach. Our JIT-based implementation shows close to theoretical peak performance, depending on the setting and the CPU architecture at hand. We additionally demonstrate how these JIT-optimized kernels can be integrated into a light-weight multi-node graph execution model. This illustrates that single- and multi-node runs yield high efficiencies and high image-throughputs when executing state-of-the-art image recognition tasks on CPUs.

关键词： convolutional codes Registers Kernel Computer architecture Optimization Lenses Computer Architecture convolutional coding Registers Lenses Kernel CPU architecture Convolution Seeds Image recognition Anatomy SIMD architecture peak performance ARCHITECTURE single instruction multiple datastream machine translation

来源：评论

学校读者我要写书评

暂无评论

Exascale Deep Learning for Climate Analytics 18

Exascale Deep Learning for Climate Analytics

引用

ACM/IEEE International Conference for High Performance Computing, Networking, Storage, and Analysis

作者： Thorsten Kurth Sean Treichler Joshua Romero Mayur Mudigonda Nathan Luehr Everett Phillips Ankur Mahesh Michael Matheson Jack Deslippe Massimiliano Fatica Prabhat Prabhat Michael Houston Lawrence Berkeley National Laboratory NVIDIA Oak Ridge National Laboratory

ISBN: (纸本)9781538683859

We extract pixel-level masks of extreme weather patterns using variants of Tiramisu and DeepLabv3+ neural networks. We describe improvements to the software frameworks, input pipeline, and the network training algorithms necessary to efficiently scale deep learning on the Piz Daint and Summit systems. The Tiramisu network scales to 5300 P100 GPUs with a sustained throughput of 21.0 PF/s and parallel efficiency of 79.0%. DeepLabv3+ scales up to 27360 V100 GPUs with a sustained throughput of 325.8 PF/s and a parallel efficiency of 90.7% in single precision. By taking advantage of the FP16 Tensor Cores, a half-precision version of the DeepLabv3+ network achieves a peak and sustained throughput of 1.13 EF/s and 999.0 PF/s respectively.

关键词： Meteorology Deep learning Training Technological innovation Computer architecture convolutional codes Computational modeling Technological Innovations Computer Architecture convolutional coding Computational modeling Meteorology Software Frameworks Functional training Extreme weather patterns single precision Training Throughput

来源：评论

学校读者我要写书评

暂无评论

Deep Cross-Modal Hashing 30

Deep Cross-Modal Hashing

引用

30th IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

作者： Jiang, Qing-Yuan Li, Wu-Jun Nanjing Univ Dept Comp Sci & Technol Collaborat Innovat Ctr Novel Software Technol & I Natl Key Lab Novel Software Technol Nanjing Jiangsu Peoples R China

ISBN: (纸本)9781538604571

Due to its low storage cost and fast query speed, crossmodal hashing (CMH) has been widely used for similarity search in multimedia retrieval applications. However, most existing CMH methods are based on hand-crafted features which might not be optimally compatible with the hash-code learning procedure. As a result, existing CMH methods with hand-crafted features may not achieve satisfactory performance. In this paper, we propose a novel CMH method, called deep cross-modal hashing (DCMH), by integrating feature learning and hash-code learning into the same framework. DCMH is an end-to-end learning framework with deep neural networks, one for each modality, to perform feature learning from scratch. Experiments on three real datasets with image-text modalities show that DCMH can outperform other baselines to achieve the state-of-the-art performance in cross-modal retrieval applications.

关键词： Neural networks Databases Training convolutional codes Binary codes Semantics Machine learning binary code convolutional coding Semantics Neural network Cross-modal Database Familial Hypertrophic Cardiomyopathy Functional training hashing Artificial Intelligence Machine Learning Training semiotics

来源：评论

学校读者我要写书评

暂无评论

Impulse Noise Suppression for Single-Carrier Coded Systems by Monte Carlo Expectation Maximization

Impulse Noise Suppression for Single-Carrier Coded Systems b...

引用

IEEE Vehicular Technology Conference

作者： Der-Feng Tseng Shi-Shun Lin National Taiwan University of Science and Technology Department of Electrical Engineering Taipei 106 Taiwan

ISBN: (纸本)9781538663592;9781538663585

An efficient approach to alleviating performance loss caused by unknown (non-stationary) impulse noise, which is typically sparse and whose statistics are difficult to model accurately, was made possible in power line communications systems when the clipping operation is properly implemented. As reported in numerous papers, the clipping threshold, which impacts system performance profoundly, is closely related with the probability of impulse occurrence and the strength of the background noise. This paper first highlights that the least-absolute-shrinkage-and-selection-operator (LASSO) algorithm, aimed at estimating the regularization parameter λ and sparse vector of regression coefficients, can be interpreted as an effective clipping operation. Subsequently, a hybrid of the Bayesian LASSO and the maximum a posteriori algorithm using the Monte Carlo expectation maximization (MCEM) is proposed to simultaneously estimate the regularization parameter and detect the convolutional codeword in single-carrier coded systems subject to the impulse noise. Moreover, by exploiting information exchangeable between those two aforementioned entities, a novel termination rule on the EM iteration stage is devised. Numerical results attest the efficacy of the proposed scheme, despite a lack of statistical knowledge on impulse noise models.

关键词： Bayes methods carrier transmission on power lines convolutional codes expectation-maximisation algorithm impulse noise interference suppression Monte Carlo methods regression analysis Impulse noise carrier transmission on power lines expectation-maximisation algorithm convolutional coding Bayes methods suppression of interference Monte Carlo technique regression coefficients regression analysis

来源：评论

学校读者我要写书评

暂无评论

Densely Connected convolutional Networks 30

Densely Connected Convolutional Networks

引用

30th IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

作者： Huang, Gao Liu, Zhuang van der Maaten, Laurens Weinberger, Kilian Q. Cornell Univ Ithaca NY 14853 USA Tsinghua Univ Beijing Peoples R China Facebook AI Res Menlo Pk CA USA

ISBN: (纸本)9781538604571

Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections-one between each layer and its subsequent layer-our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less computation to achieve high performance. Code and pre-trained models are available at https://***/liuzhuang13/DenseNet.

关键词： Training Convolution Network architecture convolutional codes Neural networks Road transportation Road transport NETWORK STRUCTURE convolutional coding Convolution Neural network Functional training Network Training FEEDFORWARD

来源：评论

学校读者我要写书评

暂无评论

On Robots Imitating Movements Through Motor Noise Prediction 7

On Robots Imitating Movements Through Motor Noise Prediction

引用

7th joint IEEE International Conference on Development Learning and Epigenetic Robotic (IEEE ICDL-EpiRob)

作者： Pico, Antonio Schillaci, Guido Hafner, Verena V. Lara, Bruno Humboldt Univ Adapt Syst Grp Berlin Germany Univ Autonoma Estado Morelos Cognit Robot Grp Cuernavaca Morelos Mexico

ISBN: (纸本)9781538637159

Robot ego-noise, that is the noise produced while the robot is moving around, can carry useful information about the motor system and the embodiment of the agent. We present an experiment where a mobile robot acquires knowledge about its ego-noise. In particular, we adopt a learning strategy based on self-exploration behaviours and on an inverse model for encoding the mappings between ego-noise and the motor commands that produced it. A convolutional Autoencoder is adopted for semi-supervised learning of auditory features. The inverse model maps both auditory features and perception of the robot speed to the motor commands that produced the ego-noise. We demonstrate how the trained models can be used for imitating movements from listening to the noise they produce.

关键词： Robot sensing systems Mobile robots Feature extraction Predictive models Convolution convolutional codes Manipulator sensing systems Mobile robots convolutional coding Robot Convolution Prediction models Feature extraction Inverse method Motor noise motor system Noise Semi-supervised Learning

来源：评论

学校读者我要写书评

暂无评论

DeepEfference: Learning to Predict the Sensory Consequences of Action Through Deep Correspondence 7

DeepEfference: Learning to Predict the Sensory Consequences ...

引用

7th joint IEEE International Conference on Development Learning and Epigenetic Robotic (IEEE ICDL-EpiRob)

作者： Shamwell, E. Jared Nothwang, William D. Perlis, Donald Univ Maryland Neurosci & Cognit Sci Program College Pk MD 20742 USA US Army Res Lab Adelphi MD 20783 USA US Army Res Lab Micro & Nano Devices & Mat Branch Sensors & Elect Adelphi MD 20783 USA Univ Maryland Comp Sci College Pk MD 20742 USA

ISBN: (纸本)9781538637159

As the human eyeball saccades across the visual scene, humans maintain egocentric visual positional constancy despite retinal motion identical to an egocentric shift of the scene. Characterizing the underlying biological computations enabling visual constancy can inform methods of robotic localization by serving as a model for intelligently integrating complimentary, heterogeneous information. Here we present DeepEfference, a bio-inspired, unsupervised, deep sensorimotor network that learns to predict the sensory consequences of self-generated actions. DeepEfference computes dense image correspondences [1] at over 500 Hz and uses only a single monocular grayscale image and a low-dimensional extra-modal motion estimate as data inputs. Designed for robotic applications, DeepEfference employs multi-level fusion via two parallel pathways to learn dense, pixel-level predictions and correspondences between source and target images. We present quantitative and qualitative results from the SceneNet RGBD [2] and KITTI Odometry [3] datasets and demonstrate an approximate runtime decrease of over 20,000% with only a 12% increase in mean pixel matching error compared to DeepMatching [4] on KITTI Odometry.

关键词： Visualization Robot sensing systems Cameras Image reconstruction convolutional codes Two dimensional displays Manipulator sensing systems convolutional coding image reconstruction Cameras Visualization target image parallel paths odometry Correspondence

来源：评论

学校读者我要写书评

暂无评论

Research and Implementation of Viterbi Decoding in TD-LTE System

Research and Implementation of Viterbi Decoding in TD-LTE Sy...

引用

2017 IEEE 第2届先进信息技术,电子与自动化控制国际会议(IAEAC 2017)

作者： Li Li-fu Li Hai-wen Li Hong-liang Gu Yong-jun Chongqing Communication Institute Zhengzhou Information Science and Technology Institute

The Viterbi algorithms of the tail biting convolutional code in TD-LTE system is introduced. Aiming at the decoding commons for the convolutional codes of TD-LTE system, this paper presents an impleme

ISBN: (纸本)9781467389808

关键词： Tail biting convolutional code Soft-decision Viterbi decoding Viterbi decoding soft decision convolutional coding Viterbi algorithm DEcoding Simulation results

来源：评论

学校读者我要写书评

暂无评论

Design of a ZigBee Based WSN with Channel coding for Energy and Temperature Monitoring

Design of a ZigBee Based WSN with Channel Coding for Energy ...

引用

Al-Sadiq International Conference on Multidisciplinary in IT and Communication Techniques Science and Applications (AIC-MITCSA)

作者： Abdullah, Hikmat N. Salman, Tariq M. Alobaidy, Haider A. H. Univ Al Nahrain Coll Informat Engn Baghdad Iraq Al Mustansiriya Univ Coll Engn Dept Elect Engn Baghdad Iraq

ISBN: (纸本)9781509032471

This paper has two objectives. First, to design and implement a prototype of WSN for monitoring power as well as temperature in buildings. Second, to evaluate the amount of improvement introduced to the ZigBee transceiver when channel-coding method is used. The results indicated that the use of ZigBee based WSN in electric power monitoring provides flexibility and helps in reducing power consumption. Furthermore, the simulation results show that using convolutional coding (CC) in ZigBee transceiver gives better performance compared to traditional ZigBee transceivers specially at CC of code rate 1/8.

关键词： Wireless Sensor Network (WSN) ZigBee IEEE 802.15.4 convolutional coding Automation Monitoring

来源：评论

学校读者我要写书评

暂无评论

Modelling the satellite communication links with orthogonal frequency-division multiplexing

引用

TRANSPORT 2016年第1期31卷 22-28页

作者： Kharchenko, Volodymir Wang, Bo Grekhov, Andrii Leschenko, Anna Natl Aviat Univ Air Nav Inst Dept Air Nav Syst Kiev Ukraine Ningbo Univ Technol Coll Econ & Management Ningbo Zhejiang Peoples R China

Global technologies of data transmission in real time are developing constantly. Therefore, modelling of messages transmission through satellite constellations is an actual problem. In this paper the original model of a communication channel 'Vehicle-to-Satellite-to-Ground Station' based on principles of the IEEE 802.11a standard is designed using MatLab Simulink software. The model allows simulating data transmission with data rate from 6 to 54 Mbit/s using adaptive modulation. Two types of channels were analysed for uplink/downlink modulation - a free space path losses with phase/frequency offset and Additive White Gaussian Noise (AWGN) link. On the base of this model, channel integrity was investigated and dependences of a Signal-Noise Ratio (SNR) on free space path losses, antennas diameters, number of OFDM symbols and satellite transponder noise temperature were received.

关键词： ground station satellite communication channel convolutional coding vehicle Viterbi decoder antenna diameter phase/frequency offset free space loss transponder noise temperature satellite transponder

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：