【摘要】 本文提出了实三层前馈网络的一种新型学习算法。该算法采用的是分层优化方法,将隐层的非线性神经元线性化,线性化产生的误差通过罚项而受到限制。分层优化使得每一层权值整体优化,而与另一层无关,这样使得整个优化过程更为有效。更多还原

【Abstract】 In this paper, a new algorithm for training of 3-layered feedforward neural network is proposed, which is based on a linearization of the nonlinear processing neurons and the optimization of the 3-layered feedforward neural networks layer by Layer. The weights in each layer are updated dependent on each other in one iteration epoch, but separately from other layers.更多还原