SupplementaryMaterialforthePaper " Sampling-DecomposableGenerativeAdversarial Recommender "

Before providing the proofs of the theorems, we restate some important notations first. In the following, denote byC the set ofN contexts, I the set ofM items andIc interacted items in a contextc. Assume G has enough capacity. If c, Sc drawn i.i.d from uniform(I), then wcj = T2) < 0. Therefore, we have P Here, we illustrate the detailed derivation of our approximated loss for learning the discriminator. We report the results on two datasets (i.e., CiteULike and Gowalla).