From: modpra@gmail.com   
      
   On Jun 20, 2:26 pm, East  wrote:   
   > I try to train BP network using basic particle swarm   
   > optimization(pso), but the result was unacceptable, the final network   
   > trained by pso have drawbacks with those trained by conventional BP   
   > algorithms: low generalization ability, unstable or badly robust,   
   > local minima and very slow convergence.   
   > any one did the work successful? or if have some source codes about   
   > pso training BP network, could send me to my email: df.n...@gmail.com   
   >   
   > any help or any suggestion is most appreciated.   
   >   
      
   The usual reason of poor generalization and non-robust back   
   propagation networks tend to be overfitting or the change in weights   
   each cycle being too high.   
   You need to tell the group what parameters you are using the PSO to   
   "evolve". Is it just the weights? or the weights and the number of   
   neurons? how many hidden layers do you have?   
   If this is a general homework problem you are trying to work out, then   
   what training data are you using? (HW problems usually have simple   
   training data like the standard Iris data set which would make   
   understanding the solution much easier)   
   Regards   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)