【作者】 郭成安； 李建华； 李明伟；

【Author】 Guo Chengan,Li Jianhua,& Li Mingwei (Dalian University of Technology)

【机构】 大连理工大学电子与信息工程学院；

【摘要】 本文以二元假设检验问题为背景,在有关观测源的先验统计知识未知的条件下,研究如何用神经网络方法从观测数据中学习后验概率函数问题。本文重点探讨神经网络模型和学习算法的设计问题,提出了一种最佳神经网络结构和强化学习算法,分析了该神经网络模型在没有理想输出值的条件下对未知的目标函数的逼近能力及其适用条件。文中还给出了计算机仿真实验结果,从实验角度验证了本文方法的有效性。更多还原

【Abstract】 With the two hypothesis detection problems as background,this paper studies how to learn the posterior probability functions from the observation data by using neural network methods under the condition where the prior statistical knowledge about the observation sources are not given in advance.A proper neural network architecture and a reinforcement learning algorithm are designed for learning the posterior probability functions.It is shown that the network architecture is an optimal model that can approach the target functions in arbitrarily small error.Simulation results given in the paper confirm the validity of the proposed approach.更多还原