基于交替方向乘子法的非光滑损失坐标优化算法.docVIP

基于交替方向乘子法的非光滑损失坐标优化算法 　　文章编号2013 　　doi：10.11772/j.issn2013.07.1912 　　摘 要： 　　交替方向乘子法（ADMM）在机器学习问题中已有一些实际应用。针对大规模数据的处理和非光滑损失凸优化问题，将镜面下降方法引入原ADMM批处理算法，得到了一种新的改进算法，并在此基础上提出了一种求解非光滑损失凸优化问题的坐标优化算法。该算法具有操作简单、计算高效的特点。通过详尽的理论分析，证明了新算法的收敛性，在一般凸条件下其具有目前最优的收敛速度。最后与相关算法进行了对比，实验结果表明该算法在保证解稀疏性的同时拥有更快的收敛速度。 　　关键词：机器学习；交替方向乘子法；坐标优化；大规模；非光滑损失 　　中图分类号：TP301 　　文献标志码：A 　　英文标题 　　New coordinate optimization method for nonsmooth losses based on alternating direction method of multipliers 　　英文作者名 　　GAO Qiankun*， WANG Yujun， WANG Jingxiao 　　英文地址（ 　　The 11th Department， Chinese Peoples Liberation Army Officer Academy， Hefei Anhui 230031， China英文摘要） 　　Abstract： 　　Alternating Direction Method of Multipliers （ADMM） already has some practical applications in machine learning field. In order to adapt to the largescale data processing and nonsmooth loss convex optimization problem， the original batch ADMM algorithm was improved by using mirror descent method， and a new coordinate optimization algorithm was proposed for solving nonsmooth loss convex optimization. This new algorithm has a simple operation and efficient computation. Through detailed theoretical analysis， the convergence of the new algorithm is verified and it also has the optimal convergence rate in general convex condition. Finally， the experimental results compared with the stateofart algorithms demonstrate it gets better convergence rate under the sparsity of solution. 　　Alternating Direction Method of Multipliers （ADMM） already has some practical applications in machine learning problem. In order to adapt to the largescale data processing and nonsmooth loss convex optimization problem， the original batch ADMM algorithm had been improved by using Mirror Descent method， and proposed a new coordinate optimization algorithm for solving nonsmooth loss convex optimization base on it. This new algorithm has a simple operation and efficient computing. Through detailed theoretical analysis，