aa0d2a804a3510442f2fd40f2100b054-AuthorFeedback.pdf

This is a strong variant of the widely used irreducibility assumption [22] (Sec. As indicated by R2, a small mini-batch BP yields a unbiased but high-variance estimate of Ex P[logΦ(x)]. In fact, nnPU slightly outperforms VPU on a learning task of "Grid Stability" as shown in Table 10.