论文部分内容阅读
为模糊形态学双向联想记忆网络(FMBAM)提出一个学习算法.在理论上证明只要存在使给定的模式对集合成为FMBAM的平衡态集合,则该学习算法总能计算出相应的最大连接权矩阵对.该最大连接权矩阵对能使FMBAM对任意输入在一步内就进入平衡态,并且神经网络全局收敛到平衡态.FMBAM的每个平衡态都是Lya-punov稳定的.当训练模式存在摄动时,利用该学习算法训练的FMBAM,对训练模式摄动拥有好的鲁棒性.
A learning algorithm for fuzzy morphological bi-directional associative memory network (FMBAM) is proposed, which theoretically proves that as long as there is a set of equilibrium states that make a given set FMBAM, the learning algorithm can always calculate the corresponding maximum weight matrix The pair of maximal connection weights enables FMBAM to enter an equilibrium state for any input in one step, and the neural network converges globally to an equilibrium state. Each equilibrium state of FAMBAM is Lya-punov stable. When the training mode exists When moving, the FMBAM trained by this learning algorithm has good robustness to training mode perturbations.