the first conception of BP neural network was brought forward in 1980s, then it was widely applied in some fields, especially in solving nonlinear questions. but, in our daily life and work we always encounter this phenomenon. if MSE is regard as criterion of the network, we aresurprised to see that the forecasting efficiency for testING data set of BP network is poor, although the training data set’s simulation results are good. how to explain this question?
The above is the main topic of discussion in this week. I‘d appreciated if you could join us.
that's a real tough question！a book is provided in order to help us get well known about this question. but I'm really stumped on how to solve this problem.
it's obviously that this question has never been addressed.
So several problems about BP neural networks must be took into consideration.
firstly, the generalization of the network is not very well, which means that the training data set is too small.
next, the network is always fall into the local minimum point. So the genetic algorithm, particle swarm optimization and etc were proposed to solve this problem.
then, stability of the network is not good.
finally,network paralysis should also be consideration.
扫码加好友，拉您进群
全部版块
我的主页
收藏
