多用户最优化的两种对偶方法

Two Kinds of Dual Methods for Multiuser Optimization

采用多用户问题的梯度近似分布式算法,对多用户最优化的原始对偶方法和正规化对偶方法进行了比较,集中于多用户凸最优化问题的概括,其中目标函数和约束函数不可分,而目标函数可通过非线性组约束,使用户决定耦合;在算法中,对原始对偶方法和正规化对偶方法可考虑不变步长,采用跨用户自然迭代计算,使每个用户能够只更新自身的决策变量.

This paper compares primal dual method and regularized dual method for multiuser optimization, focuses on the generalization of multiuser convex optimization problems, where objective function and constraint function are not separable and objective function can make users determine coupling through the constraints of nonlinear set. In order to solve this problem, gradient-approximation-distribution-style algorithm for multiuser problem is used, in this algorithm, the constant step size can be considered for primal dual method and regularized dual method and natural iterative calculation across users can be used to make each user able to only update self decision-making variables.