Hi, I've developed a feedforward backpropagation network, where I calculate the errors for that layer, and then train the weights connected to that layer, and this works fine. But I was wondering, if it would be better to calculate the errors for all layers, and then train all the weights after that has been done? Thanks.

hi. well, there are a lot of back prop algorithms on the web. just keep searching.
in my opinion there is no difference between the two variants. but i'm not an expert in nn optimization, so keep searching.

Well i tested it properly this time, it doesn't work very the new way, i just didn't realise it had to be done like that.


edit...
alls good now thanks

[Edited by - johnnyBravo on September 9, 2005 12:42:53 AM]