Minimal and maximal regularizations for tensor completion

报告摘要：We propose the minimal and maximal regularization for tensor completion. Then, the modified proximal gradient algorithm and augmented Lagrange multiplier algorithm are put forward based on their penalty forms. Meanwhile, the new potential functions are given, and the Kurdyka- Lojasiewicz property is studied. Based on these results, we obtain convergence of the proposed algorithms. Finally, we show that new regularizations and corresponding algorithms have the advantage in CPU time and recovery effect for tensor completion.

报告人简介：王川龙，太原师范学院数学与统计学院二级教授，博士生导师，1995年于西安交通大学获博士学位。曾任全国工业与应用数学学会常务理事，现任山西省工业与应用数学学会理事长、全国运筹学会理事和计算数学理事。山西省教学名师，山西省跨世纪学术和技术带头人及“三晋英才”。获山西省自然科学奖二等奖和三等奖多项，主持国家自然科学基金、山西省自然科学基金等项目10余项，在国内外学术期刊发表论文100余篇。