ce5193a069bea027a60e06c57a106eb6-AuthorFeedback.pdf

We agree (and will acknowledge more explicitly) that the overall proof program is similar to2 existingresultsinthearea. The other settings are same as those in the paper. As we can see,11 (a)showshowthe objectivevalue converges, which isalmost linear. We will like to stress that Non-linear Inductive Matrix Completion is a significantly different23 architecture than the 1-layer NNs and hence theoretical analyses for the two are quite different. Movielens dataset: our main goal in these experiments is to study the problem in theinductive28 setting,i.e.,topredictratingsfornewusers.