一种基于相对熵(CE)方法的新的随机水平值下降算法

A New Stochastic Level Value Descent Method Based on Cross-Entropy Method

为了使得随机积分水平集算法中的积分水平值能够更加有效地下降,使每次下降得到的参数更适应目标函数,本文将相对熵方法应用到随机积分水平集算法中来。利用相对熵中的ASP问题给出了一种新的参数更新方法,数值试验证明了其科学性。最后就该方法给出了更加一般的参数更新方法并给出了算法。

In order to make the integral level-value in the stochastic integral level-set algorithm more effectively decrease and the parameter better adapted to the objective function, the thesis opplies the cross-entropy method to the stochastic integral level-set algorithm. It gives a new method of parameters update by using the ASP problem, and numerical tests have shown its effectiveness; At last this thesis gives a more general method of parameter update method and the corresponding algorithm based on the ASP problem.