摘要
简述了三维转动神经网络模型,通过对能量函数的研究,证明了网络当神经元个数远大于存贮图象数时,存贮图象为能量空间中的极小值.引入了绝热近似,有效局域场能正确迭代的几率,等于方向角和旋转角分别能正确选代的几率的乘积.
After simply introducing the three-dimensional-rotational neural networks model, the energy function is studied. The energies of the stored patterns are minima in the casethe neurons number is much more larger than that of the stored patterns. With the adiabatic approximation, the probability of correct retrieval of the local field is equal to the multiplication of the probabilities of correct retrieval of the direction angle and the rotation angles.
出处
《厦门大学学报(自然科学版)》
CAS
CSCD
北大核心
1996年第3期343-347,共5页
Journal of Xiamen University:Natural Science
基金
国家自然科学基金
高技术探索资助