交替方向乘子法（Alternating Direction Method of Multipliers, ADMM）

最新推荐文章于 2024-02-26 16:36:31 发布

MadJieJie

最新推荐文章于 2024-02-26 16:36:31 发布

阅读量389

点赞数

分类专栏： Algorithms

若有帮助，请点赞&收藏，转载请标注出处。

本文链接：https://blog.csdn.net/MadJieJie/article/details/116862480

版权

本文介绍了交替方向乘子法（ADMM），它是一种用于解决优化问题的方法，尤其适用于包含两个优化变量且受约束的方程。ADMM通过耦合变量将问题转换为增广拉格朗日函数，并通过交替更新变量来迭代求解。该方法与对偶上升法相似，但能更有效地处理凸性，将大问题分解为两个子问题。

摘要由CSDN通过智能技术生成

一、 Optimization Problem

The normal optimization problem is
$min_{x} f(x)$
where the $x$ is optimization variance. By optimizing $x$ , the function $f (x)$ can be minimized.

Actually, $x$ need to be constrained by equations or inequations, such as,
$1) \ subject \ to \ {A}x=b, \ or$ $\ subject \ to \ {A}x \leq b$
Thus, the optimization problem usually can be written as

二、 Which problem can be solved by ADMM?

Optimization problems:
$\begin{array}{cc} \min_{x} f(x) +g(z) \\ Ax + Bz = c \end{array}$
ADMM algorithm usually solves optimization problems limited by equations which usually has two optimization variables.

三、 Dual Method

Dual method is the most related strategy about ADMM.
Dual method can couple/combine the minimize problem and the subject condictions by using a coupling variable $\lambda$