Changing performance function of neural network to root mean square error.

Hello,

In neural network, how can we stop the training by telling the network to stop once the root mean square error between target values and output from the network falls below say 0.01 or 0.001 ? I am specifically asking with respect to Levenberg Marquardt algorithm, such that how can we change the performance function to rmse(root mean square error)

Answer 1

[Pranav Verma]

Hi Onkar,

Please refer to the below discussion on the same lines:

https://www.mathworks.com/matlabcentral/answers/441253-when-i-should-stop-training-a-neural-network

[Onkar Khadke]

Thank you P. Verma sir. I refered to what you said. I understood how to avoid over training the network. So in trainlm the default performance function used is mse(mean squared error) and we cannot change this to rmse (root mean squared error) ?