检索结果-内蒙古大学图书馆

International Conference on Distributed Computing Systems

作者： Chang Chen Min Li Chao Yang Center for Data Science Peking University School of Mathematics Sciences Peking University National Engineering Laboratory for Big Data Analysis and Applications Peking University PKU-Changsha Institute for Computing and Digital Economy

Communication overhead is one of the major bottlenecks for large-scale distributed model training. Sparse gradient has been proposed to reduce the communication volume dramatically without affecting the model accuracy. However, high performance implementation of sparse gradient is still hindered by the overheads of gradient sparsification and inefficient implementation of sparse allreduce. For a sparse allreduce operation, the density of intermediate results may dynamically increase as summations progress. Meanwhile, high-performance sparse allreduce is further complicated with heterogeneous bandwidths. To tackle these challenges, we propose bbTopk for efficient sparse gradient training, which includes a new blocked top $k$ sparsification technique and a novel bandwidth-aware sparse allreduce algorithm. In particular, to alleviate the sparsification overhead, we design a blocked top- $k$ method, which can reduce the overhead without sacrificing model accuracy. We then build a heterogeneity-aware communication model combined with the dynamic workload feature of the sparse allreduce. Guided by that, a new sparse allreduce algorithm is proposed that can take advantage of the network resources by improving bandwidth utilization, reducing cross-node hops, and adjusting the round order. Experiments are conducted on a variety of typical neural networks and distributed environments. Results show that bbTopk can substantially outperform the previous state-of-the-art work in most test cases with up to 2.57x speedup while achieving similar accuracy with the dense model empirically.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A New Approach to the Position of Acupoints on the Back of Humans

引用

Recent Patents on Mechanical engineering 2023年第2期16卷 150-162页

作者： Xie, Shenglong Liu, Zijing Zhang, Liangan Song, Fengfeng Key Laboratory of Intelligent Manufacturing Quality Big Data Tracking and Analysis of Zhejiang Province China Jiliang University Zhejiang Hangzhou310018 China Anhui Engineering Laboratory for Intelligent Applications and Security of Industrial Internet Anhui University of Technology Anhui Ma’anshan243032 China Hangzhou Qiandaohu Ruichun Robot Research Institute Co. Ltd Zhejiang Hangzhou310018 China Zhejiang Shusi Robot Co. Ltd Zhejiang Huzhou313000 China

Aim: In order to realize the position of acupoints on the back of humans, a new acupoints positioning approach combined with machine vision technology is proposed in this paper. Background: The acupuncture point (acupoint) is an important reference in the treatment of Traditional Chinese Medicine (TCM). However, the traditional acupoints positioning method mainly depends on manpower, which has the disadvantages of low efficiency and relying on practical experience heavily. Objective: In order to realize the position of acupoints on the back of humans, a new acupoints positioning approach combined with machine vision technology is proposed in this paper. Methods: Firstly, the distribution of acupoints on the back of humans are determined according to the meridian theory, and the coordinate of each acupoint are established based on the approach of fingercun measurement, after review of previous literature and patents. Secondly, the principle of the proposed acupoints positioning method is introduced based on the skeleton model of humans. Finally, the effec-tiveness of the proposed approach is validated by experimental testing. Results: The value of emax is around 3.5mm and those of emae and erms are smaller than 0.25mm and 0.45mm, respectively. Conclusion: The experimental results prove that the proposed approach possesses the capability of good adaptive capability and high. © 2023 Bentham Science Publishers.

关键词： Computer vision

来源：评论

学校读者我要写书评

暂无评论

data Preprocessing Technology in Network Traffic Anomaly Detection

Data Preprocessing Technology in Network Traffic Anomaly Det...

引用

作者： Duan, Xueyuan Fu, Yu Wang, Kun Department of Information Security Naval University of Engineering Wuhan430033 China College of Computer and Information Technology Xinyang Normal University Xinyang464000 China Henan Key Laboratory of Analysis and Applications of Education Big Data Xinyang464000 China School of Mathematics and Information Engineering Xinyang Vocational and Technical College Xinyang464000 China

Computer network plays a very important role in today’s world. With the development of network technology and the increasing popularity of network topology, network management is facing huge challenges. Among them, finding network traffic anomalies is one of the most important tasks in network monitoring, and is an important technology in network intrusion, security monitoring, and performance maintenance. This paper aims to study data preprocessing techniques in network traffic anomaly detection. This paper firstly introduces the network traffic measurement characteristics completely, and then selects and applies the data preprocessing algorithm, constructs a complete data preprocessing framework, and conducts experiments for the abnormal detection of network multi-dimensional characteristic traffic and the abnormal detection of unit-level characteristic traffic. The superiority of the data preprocessing technology in this paper, by reducing the data, determines the data preprocessing threshold of this paper to ensure that the data preprocessing is maximized. Experiments show that the data preprocessing technology constructed in this paper has about 20% more data filtering than traditional data filtering, which shortens the amount of detection data. The threshold is around 0.6, and data preprocessing can maximize the reduction of data without destroying the integrity of the attacked data. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： data mining

来源：评论

学校读者我要写书评

暂无评论

Model Degradation Hinders Deep Graph Neural Networks

arXiv

引用

arXiv 2022年

作者： Zhang, Wentao Sheng, Zeang Yin, Ziqi Jiang, Yuezihan Xia, Yikuan Gao, Jun Yang, Zhi Cui, Bin School of CS & Key Laboratory of High Confidence Software Technologies Peking University China Center for Data Science Peking University National Engineering Laboratory for Big Data Analysis and Applications China Beijing Institute of Technology China

Graph Neural Networks (GNNs) have achieved great success in various graph mining tasks. However, drastic performance degradation is always observed when a GNN is stacked with many layers. As a result, most GNNs only have shallow architectures, which limits their expressive power and exploitation of deep neighborhoods. Most recent studies attribute the performance degradation of deep GNNs to the over-smoothing issue. In this paper, we disentangle the conventional graph convolution operation into two independent operations: Propagation (P) and Transformation (T). Following this, the depth of a GNN can be split into the propagation depth (Dp ) and the transformation depth (Dt ). Through extensive experiments, we find that the major cause for the performance degradation of deep GNNs is the model degradation issue caused by large Dt rather than the over-smoothing issue mainly caused by large Dp. Further, we present Adaptive Initial Residual (AIR), a plug-and-play module compatible with all kinds of GNN architectures, to alleviate the model degradation issue and the over-smoothing issue simultaneously. Experimental results on six real-world datasets demonstrate that GNNs equipped with AIR outperform most GNNs with shallow architectures owing to the benefits of both large Dp and Dt, while the time costs associated with AIR can be ignored. Copyright © 2022, The Authors. All rights reserved.

关键词： Graph neural networks

来源：评论

学校读者我要写书评

暂无评论

A Dynamic Acceleration Method for Remote Sensing Image Processing Based on CUDA 12th

A Dynamic Acceleration Method for Remote Sensing Image Proce...

引用

12th EAI International Conference on Simulation Tools and Techniques, SIMUtools 2020

作者： Zuo, Xianyu Zhang, Zhe Qiao, Baojun Tian, Junfeng Zhou, Liming Zhang, Yunzhou Henan Key Laboratory of Big Data Analysis and Processing Henan University Kaifeng475004 China Henan Engineering Laboratory of Spatial Information Processing Henan University Kaifeng475004 China College of Computer and Information Engineering Henan University Kaifeng475004 China National Cultural Heritage Administration Beijing100010 China

ISBN: (纸本)9783030727918

The incredible increase in the volume of remote sensing data has made the concept of Remote Sensing as big data reality with recent technological developments. Remote sensing image processing is characterized with features of massive data processing and intensive computation, which makes the processes difficult. To optimize the remote sensing image processing for GPU, compute unified device architecture (CUDA) is widely used to implement remote sensing algorithms. However, the usage of GPU in remote sensing image processing has been constrained by the complexity of its implementation and configuration. Therefore, how to take fully advantage of the parallel organization of GPU architecture is awfully challenging. In this paper, a dynamic adaptive acceleration (DAA) method is proposed to determine calculation parameters of GPU adaptively and preprocess the input remote sensing images on host dynamically. By this method, we determine calculation parameters according to the hardware parameters of GPU firstly. And then, the input remote sensing images are reconstructed based on the calculation parameters. Finally, the preprocessed image blocks are arranged to stream tasks and executed on GPU respectively. Effectiveness of the proposed DAA method in accelerate remote sensing algorithm with point operations were verified by experiments in this paper, and the experimental results indicated that the DAA method can obtain better performance than traditional methods. © ICST Institute for Computer Sciences, Social Informatics and Telecommunications engineering 2021.

关键词： Graphics processing unit

来源：评论

学校读者我要写书评

暂无评论

Mean-field theory for the Nagel-Schreckenberg model with overtaking strategy 5

Mean-field theory for the Nagel-Schreckenberg model with ove...

引用

5th International Workshop on Statistical Physics and Mathematics for Complex Systems, SPMCS 2017

作者： Su, Zhu Deng, Weibing Han, Jihui Li, Wei Cai, Xu National Engineering Laboratory for Technology of Big Data Applications in Education Central China Normal University Wuhan China Complexity Science Center Institute of Particle Physics Central China Normal University Wuhan China School of Computer and Communication Engineering Zhengzhou University of Light Industry Zhengzhou China

Based on the Nagel-Schreckenberg (NS) model with periodic boundary conditions, a modified model considered overtaking strategy (NSOS) has been proposed [1, 2]. In this paper, we focus on the theoretical analysis of traffic flow for NSOS model by using mean-field method. In the special case of vmax = 1 where vehicles can not overtake preceding ones, the features of stationary state can be obtained exactly. However, in the case of vmax > 1 where overtaking happens, some approximative methods have to be took into account. The main results are that we find the reason why traffic flow is increased in the regime where densities exceed the maximum flow density, and the influence of traffic flow on the transition density is dominated by the braking probability p. © Published under licence by IOP Publishing Ltd.

关键词： Mean field theory

来源：评论

学校读者我要写书评

暂无评论

HET: Scaling out huge embedding model training via cache-enabled distributed framework

arXiv

引用

arXiv 2021年

作者： Miao, Xupeng Zhang, Hailin Shi, Yining Nie, Xiaonan Yang, Zhi Tao, Yangyu Cui, Bin Peking University China Institute of Computational Social Science Peking University Qingdao China Tencent Inc Center for Data Science Peking University National Engineering Laboratory for Big Data Analysis and Applications China

Embedding models have been an effective learning paradigm for high-dimensional data. However, one open issue of embedding models is that their representations (latent factors) often result in large parameter space. We observe that existing distributed training frameworks face a scalability issue of embedding models since updating and retrieving the shared embedding parameters from servers usually dominates the training cycle. In this paper, we propose HET, a new system framework that significantly improves the scalability of huge embedding model training. We embrace skewed popularity distributions of embeddings as a performance opportunity and leverage it to address the communication bottleneck with an embedding cache. To ensure consistency across the caches, we incorporate a new consistency model into HET design, which provides fine-grained consistency guarantees on a per-embedding basis. Compared to previous work that only allows staleness for read operations, HET also utilizes staleness for write operations. Evaluations on six representative tasks show that HET achieves up to 88% embedding communication reductions and up to 20.68× performance speedup over the state-of-the-art baselines. Copyright © 2021, The Authors. All rights reserved.

关键词： Embeddings

来源：评论

学校读者我要写书评

暂无评论

A Heuristic Method for Route Programming in Puzzle-Based Energy Storage Systems

A Heuristic Method for Route Programming in Puzzle-Based Ene...

引用

IEEE Conference on Energy Internet and Energy System Integration (EI2)

作者： Yang Zou Jianxiao Wang School of Engineering Peking University Beijing China National Engineering Laboratory for Big Data Analysis and Applications Peking University Beijing China Peking University Ordos Research Institute of Energy Ordos China

Recently, many energy storage-related enterprises have been facing difficulties brought out by the limitation of land and the increase in loan cost. As a promising approach to improving space utilization rate, puzzle-based energy storage systems (PBESSs) are drawing more attention from logistics researchers. In previous research about PBESS, concentration has been paid to single target item problems. However, there are no consensus algorithms to solve load retrievals route programming in PBESSs with multiple target items. In this paper, a heuristic algorithm is proposed to solve load retrieval route programming in PBESSs with multiple target items, multiple escorts, and multiple (Inputs Outputs) IOs. This paper defines new concepts about the proposed algorithm, including target IOs, target position of escorts, number of escorts required, et al. Then, the decision procedures are designed according to these concepts. Based on the Markov decision process, the proposed algorithm makes the decision of the next action by analyzing the current status, until all the target items arrive at the IOs. The case study results have shown the effectiveness and efficiency of the proposed heuristic algorithm.

关键词：

来源：评论

学校读者我要写书评

暂无评论

INFORMATION GAIN PROPAGATION: A NEW WAY TO GRAPH ACTIVE LEARNING WITH SOFT LABELS

arXiv

引用

arXiv 2022年

作者： Zhang, Wentao Wang, Yexin You, Zhenbang Cao, Meng Huang, Ping Shan, Jiulong Yang, Zhi Cui, Bin School of CS Peking University China Apple United States National Engineering Laboratory for Big Data Analysis and Applications Institute of Computational Social Science Peking University Qingdao China

Graph Neural Networks (GNNs) have achieved great success in various tasks, but their performance highly relies on a large number of labeled nodes, which typically requires considerable human effort. GNN-based Active Learning (AL) methods are proposed to improve the labeling efficiency by selecting the most valuable nodes to label. Existing methods assume an oracle can correctly categorize all the selected nodes and thus just focus on the node selection. However, such an exact labeling task is costly, especially when the categorization is out of the domain of individual expert (oracle). The paper goes further, presenting a soft-label approach to AL on GNNs. Our key innovations are: i) relaxed queries where a domain expert (oracle) only judges the correctness of the predicted labels (a binary question) rather than identifying the exact class (a multi-class question), and ii) new criteria of maximizing information gain propagation for active learner with relaxed queries and soft labels. Empirical studies on public datasets demonstrate that our method significantly outperforms the state-of-the-art GNN-based AL methods in terms of both accuracy and labeling cost. Copyright © 2022, The Authors. All rights reserved.

关键词： Graph neural networks

来源：评论

学校读者我要写书评

暂无评论

Bayesian Optimization of Double-SAC Robotic Arm Path Planning Using Recent Prioritized Experience Replay 5

Bayesian Optimization of Double-SAC Robotic Arm Path Plannin...

引用

5th World Robot Conference Symposium on Advanced Robotics and Automation, WRC SARA 2023

作者： Yuan, Shuai Wen, Ke Zhang, Feng Xiong, Jianqiang Shenyang Jianzhu University Liaoning Province Big Data Management and Analysis Laboratory of Urban Construction Shenyang Branch of National Special Computer Engineering Technology Research Center School of Computer Science and Engineering Shengyang110168 China Shenyang Jianzhu University School of Electrical and Control Engineering Shengyang110168 China

ISBN: (纸本)9798350307320

As for the path planning of robotic arms, Deep Reinforcement Learning (DRL) algorithms have recently been developed. However, the high-dimensional action space and state space will result in reward sparsity and low training efficiency of robotic arm path planning. We propose a Bayesian optimization of Double-SAC algorithm using recent priority experience replay to solve these issues. We first propose the sampling method of recent prioritized experience replay (RPER), which combines recent sampling and prioritized experience replay (PER). The interval of experience pool is divided according to the recent sampling method, and sampling is performed according to the sample priority of the interval. Then we adopt a Bayesian optimization method that reduces the variance of the random strategy and improve the quality of the sampled actions. The larger variance is more beneficial for exploring the environment. Smaller variance is advantageous for exploiting good experiences to obtain higher returns. Our approach finds a good balance between exploration and exploitation. We train two Soft Actor Critic (SAC) agents to obtain two independent probability distributions, and a Bayesian optimizer is adopted to combine the two distributions to design a hybrid strategy. The hybrid strategy can output better actions according to the uncertainty estimation of the distribution, and further enhance training efficiency and stability. Finally, simulation tests are performed on the ROS-Gazebo platform, and the simulation results illustrate that the algorithm proposed in this paper converges faster and has a shorter training time than the original algorithm in terms of path planning performance. © 2023 IEEE.

关键词： Motion planning

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：