BriefGPT.xyz
Apr, 2023
降低Frank-Wolfe方法的离散化误差
Reducing Discretization Error in the Frank-Wolfe Method
HTML
PDF
Zhaoyue Chen, Yifan Sun
TL;DR
通过改进多步骤的Frank-Wolfe方法和LMO-平均方案使用一阶和高阶离散方案,从而减少离散化误差,其局部收敛速率通过一般凸集可以加速从O(1/k)到O(1/k^{3/2}),改善Frank-Wolfe算法的收敛速度。
Abstract
The
frank-wolfe algorithm
is a popular method in structurally constrained
machine learning
applications, due to its fast per-iteration complexity. However, one major limitation of the method is a slow rate of
→