Yes, if the nodes had their weights retrained. The resulting output vector from the model would subsequently have different values and would not be meaningful if used against a different set of weights.