论文部分内容阅读
利用前向网络输入元素非线性关联的方法实现了从输入模式空间到输出标识空间复杂的非线性变换．推导了学习方法，并在学习过程中把模式的平移不变识别、比例不变识别及旋转不变识别等条件构造在神经网络的权结构之中，使其具有模式不变识别能力，同时借助等权类的概念，极大地简化了网络的拓扑结构，降低了学习时间
The non-linear transformation from input mode space to output identification space is realized by using the method of non-linear input of elements in the forward network. The learning method is deduced and the conditions such as invariant translation invariant, proportional invariant invariant and invariant rotational invariant are constructed in the neural network’s weight structure in learning process so that it has the ability of pattern invariant recognition. The concept of equivalence classes greatly simplifies the topology of the network and reduces learning time