It learns the weighting value based on the difference between the ideal output and the actual output, and then modifies the weighting value via back propagation.

• 而当学习权值时,却是根据理想输出与实际的误差,由前向后逐层修改权值(Back Propagation)。