Standardising Training Set in Backpropogation

Question

If I was to standardise the training data before I train the neural network, after the training do I then de-standardise the training data and feed it back in to the neural network to show the final modelled results and expected results. Or do I feed the standardised training data back in and de-standardise the final results and expected results after?

Answer 1

You never destandarize input data. Network (or any other machine learning model) won't understand data which is in different scale/space than the one that was used during training.

If you did, however scale the output values, than in an obvious way to have to scale them back in order to obtain "unscaled" results.