首页|期刊导航|计算机与现代化|图模型数据中的子块迁移学习

图模型数据中的子块迁移学习OA

Submatrix Transfer Learning in Graphical Model Data

中文摘要英文摘要

图模型数据被广泛应用于社交网络、生物信息网络、知识图谱等领域.在多源数据分析场景中,如何充分利用并融合各数据源上的有用信息是提高图模型估计性能的关键.本文提出一种新的子块迁移学习方法,实现对多源数据之间的信息充分融合利用.首先,将节点集合分成多个互不重叠的子块,在邻域回归的基础上针对每一个源域的数据结构进行子块级筛选,利用图模型数据的精度矩阵提取各个源域数据中与目标域数据具有较高相关性的局部子块信息,聚合所提取出的可迁移子块对应的数据片段与可迁移的源域数据,共同用于目标图模型的估计,充分利用不同的数据源的信息以提高模型的泛化能力.然后,在GTEx基因组织表达数据集上,对涉及中枢神经系统神经元分化的相关脑组织数据的预测误差进行分析,最后,仿真实验结果验证了本文方法的有效性.仿真结果显示,基于子块迁移学习的方法能够充分利用多源信息,有效提高图模型数据估计的精度及稳定性,在平均相对误差等方面具有更好的性能.

Graph model data is widely used in social networks,bioinformatics networks,knowledge graphs,and other fields.In multi-source data analysis scenarios,the key to improving graph model estimation performance lies in how to fully utilize and in-tegrate useful information from various data sources.This paper proposes a novel submatrix transfer learning method that can fully integrate and utilize information from multiple data sources.Firstly,nodes are divided into multiple non-overlapping submatri-ces.Based on neighborhood regression,a submatrix-level screening method is used for the data structure of each source domain.This fully utilizes the precision matrix of the graph model data to extract local submatrix information from each source domain that has a high correlation with the target domain data.The data fragments corresponding to the extracted transferable submatrices are aggregated with the transferable source domain data and used together for the estimation of the target graph model,thus fully uti-lizing information from different data sources to improve the model's generalization ability.Subsequently,on the GTEx gene tis-sue expression dataset,the prediction error of brain tissue data related to the differentiation of neurons in the central nervous sys-tem is analyzed.Finally,simulation results are presented to verify the effectiveness of the proposed method.Simulation results show that the submatrix transfer learning method can fully utilize multi-source information,effectively improve the accuracy and stability of graph model data estimation,and have better performance in terms of mean relative error.

许大伟;周驰;丁娟

河海大学数学学院,江苏 南京 211106河海大学数学学院,江苏 南京 211106河海大学数学学院,江苏 南京 211106

信息技术与安全科学

迁移学习子块迁移学习图模型精度矩阵估计邻域回归

transfer learningsubmatrix transfer learninggraphical modelsprecision matrix estimationneighborhood regression

《计算机与现代化》 2026 (9)

29-38,117,11

中央高校基本科研业务费项目(B240201095)

10.3969/j.issn.1006-2475.2026.09.004

评论