演講公告
新聞標題: ( 2015-12-22 )
演講主題:Ensemble Algorithm for Randomized Singular Value Decomposition
主講人:陳素雲教授(中央研究院統計所)
演講日期:2015年12月29日(星期二) 下午2:30 –3:20
演講地點:(光復校區) 科學一館223室
茶會時間:當天下午2:10 (科學一館205室)
摘要內容:
The computation of singular value decomposition (SVD) is expensive when the size of the matrix becomes large. Instead of processing a full scale of the matrix, a randomized approach is to work on a much smaller matrix, which is the full matrix multiplied by a random projection matrix to a lower dimensional subspace. In this talk, we will present an ensemble algorithm to combine results from multiple randomizations. This ensemble algorithm is based on an optimization on Stiefel manifold. It consists of iterative steps of lifting orthogonal matrices (which are points on Stiefel manifold) to its tangent space, taking average on the tangent space, and then map back to the manifold. We will demonstrate the performance of the ensemble algorithm using image data. We will also give theoretic results of the asymptotic behavior of the ensemble. (joint work with David Chang, Hung Chen, Ting-Li Chen, Chen-Yao Lin and Weichung Wang)相關檔案:演講1041229-2.doc
