论文部分内容阅读
本文描述了一个在多层神经网中利用离散化正向传输,以输出均方误差的随机变量为指导,进行总体方向上权网的肯定——否定非线性学习方法。离散化正向传输减少了神经元上保持值的种类数目。总体方向上的肯定——否定简化了自学习算法。该方法摆脱了较复杂的反向传输计算而不失其学习速度。我们把这种方法称作DSGNN算法。DSGNN更适合于并行处理和生物视觉模拟。DSGNN已成功地应用于手描逻辑符线路图自学习识别系统,识别准确率在95%以上。
In this paper, we describe a positive-negative non-linear learning method that uses the random variables in the multi-layer neural network to make use of discretization of forward transmission and output of mean square error. Discretization of forward transmission reduces the number of species held on neurons. The general direction of the affirmative - negation simplifies the self-learning algorithm. This method gets rid of the complicated reverse transmission calculation without losing its learning speed. We call this method the DSGNN algorithm. DSGNN is more suitable for parallel processing and biological visual simulation. DSGNN has been successfully applied to hand-painted logic circuit diagram self-learning recognition system, recognition accuracy of more than 95%.