A Dual Entropic Regularized FS-WBP

Neural Information Processing Systems 

In this section, we provide a complete derivation of the problem in Eq. (6). By the definition of H(X), the nonnegative constraint X 0 can be neglected. In the above problem, the objective function is strongly convex. Thus, the optimal solution is unique. We specify the definition of the minimum-cost flow (MCF) problem.