学术报告
报告题目: Randomlypermuted ADMM and subspace Optimization methods
报告人：贲树军博士（中国科学院数学与系统科学研究院）
报告时间：2015年5月8日下午3:00--5:00
报告地点：4号楼4318室
欢迎广大师生参加。
数学学院
2015年5月4日
报告摘要： Thistalk is concerned with large scale optimization problems arising indata
analysis,machine learning and other areas of current interest. A popular andeasy way
todeal with these large scale optimization problems is to solve thelarge scale subproblems approximately by some certain simple methods,which aims to reduce the computation and storage cost. In this talk,I first introduce randomly permuted ADMM for these problems, which ineach step randomly and independently permutes the updating order ofany given number of blocks, and then updates the Lagrange multiplier.Then, I introduce the subspace optimization method that constructs asubproblem in low dimensions in each iteration so that thecomputation cost is reduced much more than the standard approachesdo. This offers a possible way to handle large scale optimizationproblems.