Recently I've been reading about weak supervision. I understand most of the concept details, there's one thing that is not clear to me though.

In the generative model part (creating generative model from labeling function, the Generative Model as an Expressive Vehicle section) there's a statement:

By learning a generative model, and directly estimating P(L|y), we are essentially learning the relative accuracies of the labeling functions based on how they overlap and conflict (note, we don’t need to know y!)

How is estimating P(L|y) without knowing y possible? Is this due to some assumption w.r.t. weak supervision that I didn't catch or is this related to some conditional probability property I'm not able to apply here?


1 Answer 1


Well, it seems the word note in the statement quoted didn't actually mean something obvious. Anyway, the answer is in the paper in the 2.2 Generative Model chapter.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.