报告题目:
Stochastic Gradient Descent Alternating Method for optimal Low rank Decompositions of high order sensors

报告人：曹延昭教授

报告摘要：
High order tensors have applications in many areas (biology, finance, engineering, etc). One of the key issues in high order tensor research is the optimal rank one decomposition of a tensor, which is comparable to the singular value decompositions (SVD) for matrices. Unlike SVD, the rank one decomposition problem for high order tensor is NP-hard.   The Stochastic Gradient Descent Alternating Least Squares (SALS) method is a generalization of the well-known Alternating Least Squares (ALS) method that approximates the canonical decomposition of averages of sampled random tensors. Its simplicity and efficient memory usage make the SALS   algorithm an ideal tool for decomposing tensors in an online setting. We shall show, under mild regularization and readily verifiable assumptions on the boundedness of the data, that the SALS algorithm is globally convergent.

报告时间：2020年7月12日上午9:00-12:00

报告地点：腾讯会议 ID：506 791 894

报告人简介：
曹延昭为美国奥本大学教授，主要从事偏微分方程和积分方程数值解法、随机偏微分方程数值解、非线性滤波、不确定性量化等领域的研究，部分重要研究成果发表在《SIAM J. Numer. Anal.》、《Numer. Math.》、《Math. Comp.》、《IMA J. Numer. Anal.》等计算数学国际顶尖杂志。现担任包括 计算数学国际顶尖期刊《SIAM J. Numer. Anal. 》在内的多个学术期刊编委，研究课题得到美国国家自然基金及美国空军科学研究室的长期资助。