Note that for comparable performance, you probably need not only the network topology but also the training data...