Ney man-Person准则的神经网络实现新算法

假设检验中Neyman-Person准则是一种基于似然比的信号分类、检测、识别方法. 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. 本文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-Person准则的神经网络实现新算法. 文中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Neyman-Person准则.

New Neural Network Realization Algorithm for Neyman-Person Criterion

Neyman-Person criterion in hypothesis testing is a method based on the probability rate for problems like classification, detection, and pattern recognition. Solutions through neural network to those problems would be very desirable. However, the traditional least square learning algorithms, like backpropagation, provide no guarantee for success. This paper intends to improve a kind of non-least-square learning algorithm, decide the criterion of the probability distribution and give a better algorithm based on the absolute error. Aside from theoretical argument,the proposed algorithm is examined on a simulated problem and compared with other algorithms. The simulative result proves that the new algorithm has fewer errors and is more suitable for the Neyman-Person criterion.