This algorithm determines the step which revises the connecting data between neural units by the learning principle that the input studying sample can produce the mast limited network error.

• 该算法以输入的学习样本产生的网络误差下降最大的学习原则确定对神经元间连接权值修正的迭代步长。