How to Apply Deep Learning to a Problem Like the Right-Sharing Problem

When you're approaching a problem like the right-sharing problem, how do you transform this into something that you can apply directly deep learning or RL to? So effectively, you need to build up a model of the environment, first, which comes from the real world. That's exactly the same method that we would use, for example, on a problem like deenpacking. Usually, on NP-hard problems in logistics, it's very easy to verify if a given solution is a good one or not.