I am working on building on building a black box system as part of my master's thesis.
I am wondering if anyone has any ideas of a novel method I could pursue that I could do in this realm that I could write a master thesis about?
BTW, that freeze thaw idea is really great!
I am looking for something new for my thesis that could be publishable.
One idea I had was creating a database of (dataset, feature selection method, model used, hyperparameters, score) that could say something like "if model X works well for a dataset, than feature transformation Y would likely also work well", like meta-learning done by the auto-sklearn team.
Also, I would be glad to collaborate with another team implementing meta-learning ideas if they want any help :).