439d8c975f26e5005dcdbf41b0d84161-AuthorFeedback.pdf

Regarding the notion of estimating learnability more generally, although it is still relatively new, there is already13 a growing body of work (appearing at recent NeurIPS, COLT, and AISTATS conferences; see lines 41-48 of our14 submission for references), studying itfor avariety oflearning problems. The focus ofour work ison formal performance guarantees, and such21 guarantees for top-down algorithms are only known for monotone target functions.