1: \begin{abstract}

2: Approximations of loopy belief propagation, including expectation propagation

3: and approximate message passing, have attracted considerable attention for

4: probabilistic inference problems.  This paper proposes and analyzes

5: a generalization of Opper and Winther's

6: expectation consistent (EC) approximate inference method.

7: The proposed method, called Generalized Expectation Consistency (GEC),

8: can be applied to both maximum a posteriori (MAP) and minimum mean squared error (MMSE)

9: estimation.

10: Here we characterize its fixed points, convergence,

11: %---a key challenge in belief propagation methods---

12: and performance relative to the replica prediction of optimality.\!

13: \end{abstract}

14: