A linearly convergent stochastic recursive gradient method for convex optimization
报告人:王晓 副教授, 中国科学院大学 时间: 2019年10月25日16:20
报告地点:行健楼-526
邀请人:姜波 副教授
摘要:The stochastic recursive gradient algorithm (SARAH) attracts much interest recently. It admits a simple recursive framework for updating stochastic gradient estimates. Motivated by this, in this paper, we propose a SARAH-I method incorporating importance sampling, whose linear convergence rate of the sequence of distances between iterates and the optima set is proven under both strong convexity and non-strong convexity conditions. Further, we propose to use the Barzilai-Borwein (BB) method to automatically compute step sizes for SARAH-I, named as SARAH-I-BB, and we establish its convergence and complexity properties in different cases. Finally numerical tests are reported to indicate promising performances of SARAH-I-BB.