论文部分内容阅读
通过分析前馈神经网络中各层权系数与误差能量之间的关系,在服从最小扰动原理下,本文提出了一个新的学习方法。该方法将网络训练问题变换为一系列的凸规划子问题,而这些子问题都可以在较短时间内获得全局最优解。文中给出的计算结果表明该方法很有发展前景。
By analyzing the relationship between the weights and error energies of feedforward neural networks, this paper presents a new learning method under the principle of minimum disturbance. This method transforms the network training problem into a series of convex programming sub-problems, and these sub-problems can obtain the global optimal solution in a short time. The calculation results given in this paper show that this method is promising.