[eternal_virgin]

It would be great to have some sort of roadmap depending on "what" the user is trying to do.

i.e. if I was a total noob and I wanted to make an application using AI to detect if people in the crowd were bored, I would have no idea where to start without reading/researching for hours online on different fields and models that work and how they work. It would be neat if there was a tool that just asked you a few questions, then took that info and gave you a roadmap, i.e. "Feed Forward Neural Networks, Digit Classification, Image Classification w/ Inception, Object Detection with ResNet + Inception, Optimizing TensorFlow code for Servers, Deploying TensorFlow with Docker, Protecting Against Adversarial Input"

This way someone with a time sensitive project doesn't have to learn TF for 6 months before being able to accomplish what they wanted! Just something I think would be neat and also possible to add to TF World.

[jorgemf]

There are not so many type of problems for such a complex tool. Your problem usually fits in one category among classification, prediction, clustering, generation or control. Then you have different domains as images, video, audio, text, etc. With a combination or type of problem and domain you sure can have a roadmap, but you probably will need to read papers to solve your problem if it is not something some has done before.

[Aduket]

What do you mean by “control”? Like control engineering? How is it used in control domain, can you explain a bit?

[madmadjo]

+1 for this one!