Bayesian Tensor Decomposition with Sparse Block Modeling for Data Recovery

发布者:张译文发布时间:2025-04-30浏览次数:10

江苏省应用数学(中国矿业大学)中心系列学术报告

报告题目:Bayesian Tensor Decomposition with Sparse Block Modeling for Data Recovery

报告人:骆其伦 特聘研究员 华南师范大学数学科学学院

报告时间:202558日(周四)11:00-12:00

报告地点:数学A321

报告人简介:骆其伦,分别于2015年和2018年在华南师范大学获学士学位和硕士学位,并于2022年在美国伊利诺伊州南伊利诺伊大学卡本代尔分校获数学博士学位。目前是华南师范大学特聘研究员,主要研究方向为数值代数、张量分解与应用、图像处理。学术成果在IEEE TNNLSIEEE TCIEEE TSPACM TISTPattern Recognit.SIAM J. Imaging Sci等多个学术刊物上发表。主持国家自然科学基金青年项目与广东省自然科学基金面上项目各一项

报告摘要:This talk presents a Bayesian tensor decomposition framework that integrates sparse constraints with structural block decomposition via the matrix outer product. By enforcing sparsity within the latent factor tensor, the proposed model enhances the representation of spatial correlations in large-scale multidimensional datasets. Our approach is particularly effective in recovering underlying data structures in applications such as traffic data imputation. Experimental evaluations demonstrate that this methodology not only improves reconstruction accuracy but also maintains computational efficiency.