基于动态网络嵌入的结构感知图划分研究
结题报告
批准号:
62002226
项目类别:
青年科学基金项目
资助金额:
24.0 万元
负责人:
李琪
依托单位:
学科分类:
计算机系统结构与硬件技术
结题年份:
2023
批准年份:
2020
项目状态:
已结题
项目参与者:
李琪
国基评审专家1V1指导 中标率高出同行96.8%
结合最新热点,提供专业选题建议
深度指导申报书撰写,确保创新可行
指导项目中标800+,快速提高中标率
客服二维码
微信扫码咨询
中文摘要
随着来自交通、信息、金融等领域的大型动态图数据集的爆炸式增长,图划分对分布式图计算性能的影响越来越重要。本项目围绕基于动态网络嵌入的结构感知图划分中若干关键科学问题进行研究:(1)考虑集群节点计算力、网络带宽的异构和多核节点内共享资源的竞争,提出集群异构与共享资源竞争体系结构的任务流模型,实现底层体系结构指导计算资源的分配;(2)研究基于动态网络嵌入的图划分模型,设计损失函数使模型能够动态学习网络结构的更新表示,结合体系结构量化模型,实现优异的划分效果与感知结构的目的;(3)考虑到动态网络环境下生成图结构的突变性以及主从节点的任务类型不同,提出面向“主节点”的基于邻边的动态缓存存储结构和面向“从节点”的基于红黑树边的分布式邻点存储结构,能够提升图划分效率的同时,更好地支持具有突变性质的图算法。研究成果将为构建结构感知的分布式图分析系统提供理论基础与技术支撑,具有重要科学意义。
英文摘要
With the explosive growth of dynamic large-scale graph datasets from various fields from transportation, information, finance and other fields, graph partition has an increasingly important impact on the performance of distributed graph computing. This project mainly studies several key problems in graph partition of architecture-aware based on dynamic network embedding. First, considering the heterogeneous of compute ability and network bandwidth of cluster nodes, and the competition of shared resources within multi-core nodes. We will propose a task flow model of cluster heterogeneous and shared resource competition architecture, and realize the allocation of computing resources guided by the underlying architecture. Second, after studying the graph partition model based on dynamic network embedding. We will design a loss function to enable the model to dynamically learn the updated representation of network structure. Combined with the quantitative model of architecture, the excellent partition effect and the purpose of perceptual architecture will be realized. Finally, Considering the mutation of graph structure and the different task types of master-slave nodes in dynamic network environment. We will propose a dynamic cache storage structure based on adjacent edge for master node and a distributed neighbor storage structure based on red black tree edge for slave nodes, which can improve the efficiency of graph partition and support graph algorithm with mutation better. The research results will provide the theoretical basis and technical support for the construction of architecture-aware distributed graph analysis system, which is of great scientific significance.
期刊论文列表
专著列表
科研奖励列表
会议论文列表
专利列表
DOI:10.1038/s41598-023-33324-7
发表时间:2023-04-25
期刊:SCIENTIFIC REPORTS
影响因子:4.6
作者:Li, Qi;Chen, Wenping;Fang, Zhaoxi;Ying, Changtian;Wang, Chen
通讯作者:Wang, Chen
DOI:10.3390/math11092175
发表时间:2023-05
期刊:Mathematics
影响因子:2.4
作者:Huawen Liu;Minhao Yin;Zongda Wu;Liping Zhao;Qi Li;Xinzhong Zhu;Zhonglong Zheng
通讯作者:Huawen Liu;Minhao Yin;Zongda Wu;Liping Zhao;Qi Li;Xinzhong Zhu;Zhonglong Zheng
DOI:--
发表时间:2023
期刊:Digital Communications and Networks
影响因子:7.9
作者:Yizhou Shen;Shigen Shen;Qi Li;Haiping Zhou;Zongda Wu;Youyang Qu
通讯作者:Youyang Qu
DOI:10.1016/j.aeue.2023.154820
发表时间:2023-07
期刊:AEU - International Journal of Electronics and Communications
影响因子:--
作者:Zhaoxi Fang;Qi Li;Jianhua Liu;Jin Zhou;Shigen Shen
通讯作者:Zhaoxi Fang;Qi Li;Jianhua Liu;Jin Zhou;Shigen Shen
DOI:10.12263/DZXB.20190919
发表时间:2021
期刊:电子学报
影响因子:--
作者:柳菁;李琪
通讯作者:李琪
碎片化数据驱动的知识图谱动态强化方法研究
  • 批准号:
    LY22F020003
  • 项目类别:
    省市级项目
  • 资助金额:
    0.0万元
  • 批准年份:
    2021
  • 负责人:
    李琪
  • 依托单位:
国内基金
海外基金