A Approximate Marginal Inference Algorithm Algorithm 2 C-GBP: Convex Generalized Belief Propagation [25 ] Input: Region Graph G = (V, E), parameters θ = (θ > 0 Output: Model marginals τ = (τ