CAJ | 학술논문

利用三角模的模糊联想记忆网络的性质以及模糊联想记忆网络的鲁棒性定义，对基于爱因斯坦t-模构建的模糊双向联想记忆网络的学习算法的全局鲁棒性进行了分析。从理论上证明了当训练模式的摄动为正向摄动时，该学习算法可以保持良好的鲁棒性，并用实验验证了该结论；而当摄动存在负向波动时该学习算法不满足全局鲁棒性。然后又进一步对训练模式集摄动最大摄动与输出模式集的最大摄动之间的关系进行研究，得出了训练模式集的最大摄动与输出模式集的最大摄动之间的关系曲线。
이용삼각모적모호련상기억망락적성질이급모호련상기억망락적로봉성정의，대기우애인사탄t-모구건적모호쌍향련상기억망락적학습산법적전국로봉성진행료분석。종이론상증명료당훈련모식적섭동위정향섭동시，해학습산법가이보지량호적로봉성，병용실험험증료해결론；이당섭동존재부향파동시해학습산법불만족전국로봉성。연후우진일보대훈련모식집섭동최대섭동여수출모식집적최대섭동지간적관계진행연구，득출료훈련모식집적최대섭동여수출모식집적최대섭동지간적관계곡선。
The paper analyses the robustness of learning algorithm for fuzzy associative memory based on Einstain’s t-norm by using the properties of fuzzy bidirectional associative memories based on triangular norms and the overall situation robustness of fuzzy bidirectional associative memories. The conclusion that the learning algorithm can keep good overall robustness when the perturbations are positive is proved in theory and verified by experiment in this paper. And that the learning algorithm doesn’t satisfy overall situation robustness when the noise contains negative value is proved by experi-ment. What is more, the relation between the maximum of perturbations of training patterns and the maximum of perturba-tions of the output is also analyzed and the relation curve is gotten.