Then the demo prepares training by setting up a loss function (binary cross entropy), a training optimizer function (stochastic gradient descent), and parameters for training (learning rate and max ...