分离错误最小化的极大熵方法

A Maximum Entropy Method for Misclassification Minimization

分离错误最小化是支持向量机的基本问题之一,一种形式是最小化分离错误点的偏离和,这是一个不可微优化问题,笔者提出用极大熵函数将其转化成可微凸规划问题来处理,得到原问题的近似最优解.

Misclassification Minimization is a fundamental problem of machine learning. It can be stated by a way of minimizing the sum of violations of misclassified points. It is NP-complete. The objective function is not differentiable. In this paper,a convex entropy function is used to solve the nondifferentiable optimal problem,and the approximate solution is achieved by this convex programming.