Creating deep neural networks with 3 to 5 lines of code

When dealing with supervised learning within deep learning, we might say that there are some classical approaches to follow. The first solution is the so-called "heroic" strategy where one creates a completely new deep neural network (DNN) from scratch and train/evaluate it. In practical terms, this solution may not be very interesting since there are countless DNNs available nowadays, like many deep convolutional neural networks (CNNs), that can be reused. The second path is simply to consider a deployable DNN, trained for a certain context, and see its operation in another context. Despite all the advances in deep learning, models can present bad performances if the contexts are too diverse.