I have 2 CNN models (M1 and M2), which are trained simultaneously. I want the models will mutually learn from each other so that If model M1 makes a correct classification, a mistake made by the M2 leads to a synergic error that serves as an extra force to update the model.