Split Method of Multipliers and Its Application to Parallel and Distributed Logistic Regression

作     者：FENG Hui TANG Xianding YANG Tao HU Bo 

作者机构：Department of Electronic Engineering Fudan University 

出 版 物：《Chinese Journal of Electronics》 (电子学报(英文))

年 卷 期：2014年第23卷第2期

页      面：305-310页

核心收录：

学科分类：0711[理学-系统科学] 07[理学] 08[工学] 080401[工学-精密仪器及机械] 0804[工学-仪器科学与技术] 080402[工学-测试计量技术及仪器] 

基　　金：supported by the National Science and Technology Major Project of China(No.2012ZX03001007-003) 

主　　题：Method of multiplier Primal-dual method Parallel optimization Distributed optimization Logistic regression 

摘      要：We consider the scenario where two variables need to be optimized simultaneously. The minimization over one variable has an analytical solution, while it is intractable for the other. Under the Lagrangian dual framework, we propose two iterative optimization algorithms, which make partial minimization and gradient descent alternatingly over two variables. The first algorithm asserts that the iteration result converges to a KKT point under proper stepsize rules, which only needs the augmented Lagrangian function to be convex over partial variable. The second algorithm provides the local attraction property around the KKT point. Our algorithms provide a general solution to parallel and distributed optimization with summable objective functions. Simulation results on parallel and distributed logistic regression classification are present, which show faster convergence rate with less computational complexity compared with other methods.