一种多层神经网络的学习算法

A LEARNING ALGORITHM FOR A MULTILAYERED NEURAL NETWORK

提出一种基于序贯预测误差方法（ＳＰＥ）的多层神经网络（ＭＮＮ）的学习算法。经模拟计算，它比传统的基于最陡下降方法的误差反传（ＳＤＢＥＰ）算法具有更好的收敛性能。并对这两种算法进行了模拟计算的比较．

In this paper, a learning algorithm based on the sequencial predict error method (SPE) for a multilayered neural network (MNN) is derived. Simulation computations show that it converges faster than the conventional steepest descent backwards error propagation (SDBEP) algorithm. Comparison of these two algorithms is also made through simulation computations.