In Python, currently we have libraries like Theano, TensorFlow and PyTorch where you can write low level code for Deep Learning, then there are libraries like Keras (built on top of Theano and Tensorflow) and Lasagne (built on top of Theano) which gives us more ease to create a production ready model.
We have tiny-dnn in C++ which has shown very promising speed up on CPU.
But any how the user has to write code in one of the language and it hinders a person's reach to Deep Learning model. When nearly every Deep Learning Paper has a graphical representation of the model, and nearly every Deep Learning library creates a graph of the model before training, why can't we just let the user create a graph and run the model?
But there is a compromise, the more high level abstraction you get to manipulating with the model becomes harder, but most of the deep learning model in production needs no manipulation as such.
We propose to create an interface such that the user would be able to create a graph without writin