Yes, but in some fashion, all the outputs are mathematical convolutions of the training, plus whatever is packed into context.