摘要
在绝热近似基础上，应用信噪比理论分析了网络的存贮容量．当神经元的旋转角态数和方向角态数都很大时，存储容量与神经元的旋转角态数的平方成反比，和方向角态数成正比．当神经元态数较小时，结果趋于Ｈｏｐｆｉｅｌｄ神经网络的存储容量．
By using the theory of adiabatic approximation, the storage capacity of the three-dimensional rotation neural network is estimated by the signal-to-noise theory. The storage capacity is proportional inversely to the square of the rotation angles states and to the direction angles states of neurons when the neuron states is very big, and approaches to that of the Hopfield model when the neuron states is small.
出处
《厦门大学学报（自然科学版）》
CAS
CSCD
北大核心
1996年第6期862-866,共5页
Journal of Xiamen University：Natural Science
基金
国家自然科学基金
关键词
神经网络
三维转动
图象识别
存贮容量
Neural network, Three-dimension rotation, Pattern recognition, storage capacity
