寻找鞍点 minxmaxyf(x,y)\min_{x}\max_{y} f(x,y)minxmaxyf(x,y)的一个著名算法
输入:
(α,ϵ)(\alpha,\epsilon)(α,ϵ)-relaxed proximal oracle O(z)\mathcal{O}(z)O(z) for gradient mapping ggg, distance generating rrr.
参数:
迭代次数KKK
输出:
点zˉK\bar{z}_{K}zˉK,满足E[Gap(zˉ)]≤αΘK+ϵ\mathbb{E}[Gap(\bar{z})]\leq \frac{\alpha\Theta}{K}+\epsilonE[Gap(zˉ)]≤KαΘ+ϵ.
初始化:
z0=argminr(z)z_{0}=\arg\min r(z)z0=argminr(z)
for k=1,2,...,Kk = 1,2,..., Kk=1,2,...,K do:
zk−1/2=O(zk−1)\quad z_{k-1/2}=\mathcal{O}(z_{k-1})zk−1/2=O(zk−1)
zk=Proxzk−1α(g(zk−1/2))\quad z_{k}=Prox^{\alpha}_{z_{k-1}}(g(z_{k-1/2}))zk=Proxzk−1α(g(zk−1/2))
return zˉK=1K∑k=1Kzk−1/2\bar{z}_{K}=\frac{1}{K}\sum_{k=1}^{K}z_{k-1/2}zˉK=K1∑k=1Kzk−1/2