【作者】 吕迎阳； 吴伯僖； 徐慎初； 陈振湘； 陈剑勇；

【Author】 Yingyang Lu,Boxi Wu,Shenchu Xu,Zhenxiang Chen,Jianyong Chen Department of Physics,Xiamen University,Xiamen,Fujian,361005

【机构】 厦门大学物理系；

【摘要】 Jianyong chen等人提出以混沌序列初始化BP神经网络的连接权重及阈值有助于提高网络的学习成功率,并以对异或问题的学习作为例证。本文从几何的角度对其例证给予解释,并认为混沌初始化是否有利于网络的学习仍需进一步研究.更多还原

【Abstract】 Jianyong Chen et cl had proposed that initializing the connection weights and thresholds of BP neural network with chaotic values can improve the success rate of learning,and given the XOR problem as an example.In this paper we explain their example with geometric knowledge,and think that more studies are needed to say whether chaotic initial values can favor the learning.更多还原