假设检验中Neyman-Pearson准则的神经网络实现新算法

New neural network realization algorithm for Neyman-Pearson criterion in hypothesis testing

假设检验中Neyman Pearson准则是一种基于似然比的信号检测、识别、分类方法 .神经网络是实现这种判定准则的优选方案 ,但是传统的最小平方学习算法 ,如BP算法等 ,往往不能取得全局最优解 .针对一种非最小平方学习算法 ,提出了一种概率分配原则 ,并给出了一种Neyman Pearson准则的神经网络实现新算法 .对新算法在假设检验中的应用进行了仿真验证 .结果表明新算法具有更小的误差 ,更加适用于Ney man Pearson准则 .

The Neyman-Pearson criterion in hypothesis testing is based on the probability rate for problems such as classification, detection, and pattern recognition as an improved kind of non-least-square learning algorithm to decide the criterion of the probability distribution. An algorithm based on the absolute error is given. Simulation results show that the new algorithm has less error and is more suitable for Neyman-Pearson criterion.