基于内积方向的神经网络学习算法及应用

A Neural Network Learning Algorithm Based on Direction of Inner Product and Its Application

对二进前向感知器各神经元的样本空间进行了分析，利用其内积特性及吸引域概念提出一种快速、可靠、实用的学习算法．通过阈值设置与内积方向相联系，使神经网络具备可控制的容错能力，此神经网络结构简单，容易用硬件实现．通过实例说明了这种方案应用于模式分类、布尔函数逼近的途径及优良的性能．

First analyzed is the pattern space from each neuron of binary feedforward perceptron. Then a rapid, reliable and available learning algorithm using the characteristics of the inner product and area of attraction. In relation to threshold of neuron with pattern direction of the inner product, the neural network is trained to have capacity of controllable error correction,with simple structure of the neural network and easiness to be implemented by hardware. Finally this strategy is demonstrated to have good performance when used for pattern recognition and Boolean function approximation.