基于正则化风险函数模型的束方法

The Bundle Methods Based on the Regularized Risk Minimization Model

摘要: 本文在机器学习(ML)的正则化风险函数模型(SRM)的基础上，结合求解非光滑函数的束方法，提出了一种求解经验风险函数模型的算法，用割平面模型去近似目标函数，用非精确线搜索去获得步长，在适当的假设下，分析了算法的全局收敛和收敛速度。

Abstract: Based on the regularized risk function model (SRM) of machine learning (ML), this paper combines the bundle method for solving non-smooth functions, presents an algorithm for solving empirical risk function model. The objective function is approximated by the cut-plane model, and the step size is obtained by the inexact line search. Under appropriate assumptions, the global convergence and convergence speed of the algorithm are analyzed.