报告地点：行健楼学术活动室665
邀请人：蔡邢菊教授
报告摘要：In this paper, we focus on exploiting the inherited low-rank structure of a non-negative and sparse matrix based on the rectified linear unit(ReLU) activation function. We explore the ReLU-based regularized matrix decomposition model and introduce an accelerated alternating partial Bregman proximal gradient method (AAPB) for solving it. Under mild assumptions, we demonstrate the sublinear convergence and global convergence of the proposed algorithm and provide closed-form solutions for several regularizations by carefully choosing the kernel generating distance. The advantage of our algorithm is that the smooth adaptable constant only needs to be computed once, and some variables can be updated in parallel. Numerical experiments on synthetic and real datasets confirm the effectiveness of our model and algorithm.
报告人简介：
韩德仁，教授，博导，现任北京航空航天大学网赌
院长、教育部数学类专业教指委秘书长。2002年获南京大学计算数学博士学位。从事大规模优化问题、变分不等式问题的数值方法的研究工作，发表多篇学术论文。曾获中国运筹学会青年运筹学奖、江苏省科技进步奖等奖项，主持国家自然科学基金重点项目等多项。担任中国运筹学会常务理事、中国运筹学会数学规划分会常务理事、青年工作委员会主任，《数值计算与计算机应用》《Journal of the Operations Research Society of China》《Journal of Global Optimization》编委。
