Which one is more important? MSE or input-error correlation?

I have been training the neural network however I have some questions regarding the results. For some results, the mean squared error is small (for example, 5e-11) but the input-error correlation is very high (all falls under the confidence limit) do I take those as satisfactory? Also, sometimes, for two trained networks, one has a smaller MSE but larger input-error correlation, which one should I pick?

After training the neural network, I saved the advanced scripts as .m files. From there, how do I input new set of inputs to predict the new target?

Thanks!

Answer 1

[Greg Heath]

The objective is to minimize MSE.

A high input/error Xcorr indicates that the MSE can be lowered by modifying the design.

ynew = net(xnew) will predict a new output (NOT a new target...targets are known quantities).

Hope this helps.

Thank you for formally accepting my answer

Greg