I concur. It might not be feasible in terms of computational power available, but I don't think there is anything fundamentally stopping application of those training mechanisms, unless the whole neuralnet paradigm is fundamentally incompatible with those learning methods.