题 目：Representation Based Subspace Learning: Theory, Algorithm and Application
Subspace learning has many applications in real world, such as motion segmentation, image recognition, hyper-spectral remote sensing, multi-label learning, network topology discovery and so on. In these applications, data points canbe viewed as samples drawn from several unknown subspaces. The fundamentalproblem of these applications is how to partition given data points into severalsegments such that each segment contains points belonging to the same subspace,and furthermore, recover the unknown subspaces by the segments. Many algorithms have been proposed to solve this problem in last decades. Among them,the representation based methods attracted most attention because of their state-of-art performance. However, if the subspaces are complicatedly intersected orthe data points are heavily noised, existing representation based methods also fail in both empirical performance and theoretical guarantee. This report gives a detail discussion on the progress we have madeabout the above issue.
夏雨晴，大数据专业讲师。毕业于浙江大学，新加坡国立大学博士后。主持浙江省自然科学基金1项。曾在Journalof machine learning research、IEEE signal processing letters、计算数学等国内外高水平期刊发表多篇论文。