Confronting Domain Shift in Trained Neural Networks