CAJ | 학술논문

针对现有人工神经网络学习新知识会破坏已获得知识的问题,根据生物智能扩展的思想,提出一种粒计算模糊增殖神经场学习方法,将粒计算商空间理论和人工神经场理论有机结合,并引入自治神经网络中,采用分治方法和嵌入机理,把大任务分成小任务,实现知识积累、继承和不断完善.实验结果证明了该方法的合理性和可行性.
침대현유인공신경망락학습신지식회파배이획득지식적문제,근거생물지능확전적사상,제출일충립계산모호증식신경장학습방법,장립계산상공간이론화인공신경장이론유궤결합,병인입자치신경망락중,채용분치방법화감입궤리,파대임무분성소임무,실현지식적루、계승화불단완선.실험결과증명료해방법적합이성화가행성.
In view of the problem that the existed artificial neural network will destroy the obtained knowledge when it studies new knowledge, according to expansion idea of the biological intelligence,a learning method of Fuzzy Incremental Neural Field Based-on Granular Computing(FINFGC) is proposed. This method organically combines the character of the quotient space with the theory of neural field and introduces in the autonomous neural network,makes use of the partitioning method and the inserting mechanism,divides the big task into the small task so as to realize the knowledge accumulation, inheritance and perfection unceasingly. Experimental results show this method is rational and feasible.