Uber just released recently a new library called Ludwig that allows you to combine any kind of input (text, images, tabular data, time series) with an encoder and then output any kind of output (text, single label, multi label, regression etc). In terms of the vision for fastai of making deep learning super accessible even for non-coder, I thought the Ludwig API was inspiring. They have a yaml way of specifying the model and also a python API.
As a newcomer in the world of deep learning, what I found interesting with fastai is that fastai is super strong for transfer learning and have various well defined tasks where it is good at. Where it is harder is combining various kind of data together like Ludwig allows. I think it would unlock a lot of scenarios users try to do and re-using pre-trained models for image and text input could be really interesting.
Thanks for the great library guys,