神经网络中将任务学习与模型学习相结合的E-H和E-H-W学习方法

本文从获取好的神经网络泛化能力出发,首先提出了将Hebbian学习与增加问题复杂性统一起来的思想,并通过在总的误差函数中增加一限制函数来实现Hebbian学习.基于此,提出了一种将误差驱动的任务学习与Hebbian规则的模型学习相结合的E-H方法.然后,根据模型学习应同时考虑减小网络复杂性和增加问题复杂性的思想,又提出了一种将误差驱动的学习与Hebbian规则、简单的权退化法结合起来,共同来提高神经网络的泛化能力的E-H-W方法.最后通过大量实例仿真将它们与纯误差驱动的方法、权退化法、其它文献中的相关方法进行了比较.结果表明我们的方法具有最好的泛化能力,是很有效的神经网络学习方法.

E-H and E-H-W Learning Algorithms in Neural Network Combining Tasking Learning and Model Learning

In order to obtain better generalization capability of neural network,the paper first proposes an idea of uniting Hebbian rule with increasing problem complexity,and implements Hebbian rule by incorporating a functional constraint into overall object function.Based on this,a neural network learning algorithm,E H,is proposed,which combines error driven' s task learning and Hebbian rule' s model learning.Then,according to the idea of that model learning should simultaneously consider decreasing network complexity and increasing problem complexity,a new learning algorithm,E H W,is proposed,which combines error driven learning,Hebbian rules,and the simple weight decay.Finally,several example simulations are made to compare our algorithms with traditional error back propagation,simple weight decay,and relative methods in other paper.Simulation results are given to verify the effectiveness of the proposed learning algorithms.The results indicate that our learning algorithms have best generalization capability.