提出一种前向神经网络只有一个隐元就可以解决奇偶校验或对称性校验问题,讨论了如何根据输入模式的先验知识来推导隐元的传递函数和构造该分类神经网络的方法.和已有的研究工作相比,提出的前向神经网络是解决某些线性不可分的模式识别问题的最简单的网络结构,所提出的分析方法也可推广用于类似的网络构造问题. Proposed a forward neural network can solve the problem of parity check or symmetry check by only one hidden element, and discussed how to derive the transfer function of hidden element according to the prior knowledge of input pattern and how to construct the neural network. Compared with the existing research work, the proposed forward neural network is the simplest network structure to solve some linearly inseparable pattern recognition problems. The proposed analysis method can also be generalized for similar network construction problems.