I am trying to train my recurrent network using Levenberg Marquardt(LM). I do it in Matlab. I am sure my inputs and outputs are ok.
My net have 4 layer (20,2,20,5)with some feedbacks in itself.
During training, the mu increases aggresively abruptly and trainging process stops with following message:
Max Mu reached.
Is there anybody knows why the mu increases in LM up to 1e22 after 3 iteration?
Any comment will be helpfull.