English

Inference for Generalized Linear Models via Alternating Directions and Bethe Free Energy Minimization

Information Theory 2016-05-03 v2 math.IT

Abstract

Generalized Linear Models (GLMs), where a random vector x\mathbf{x} is observed through a noisy, possibly nonlinear, function of a linear transform z=Ax\mathbf{z}=\mathbf{Ax} arise in a range of applications in nonlinear filtering and regression. Approximate Message Passing (AMP) methods, based on loopy belief propagation, are a promising class of approaches for approximate inference in these models. AMP methods are computationally simple, general, and admit precise analyses with testable conditions for optimality for large i.i.d. transforms A\mathbf{A}. However, the algorithms can easily diverge for general A\mathbf{A}. This paper presents a convergent approach to the generalized AMP (GAMP) algorithm based on direct minimization of a large-system limit approximation of the Bethe Free Energy (LSL-BFE). The proposed method uses a double-loop procedure, where the outer loop successively linearizes the LSL-BFE and the inner loop minimizes the linearized LSL-BFE using the Alternating Direction Method of Multipliers (ADMM). The proposed method, called ADMM-GAMP, is similar in structure to the original GAMP method, but with an additional least-squares minimization. It is shown that for strictly convex, smooth penalties, ADMM-GAMP is guaranteed to converge to a local minima of the LSL-BFE, thus providing a convergent alternative to GAMP that is stable under arbitrary transforms. Simulations are also presented that demonstrate the robustness of the method for non-convex penalties as well.

Keywords

Cite

@article{arxiv.1501.01797,
  title  = {Inference for Generalized Linear Models via Alternating Directions and Bethe Free Energy Minimization},
  author = {Sundeep Rangan and Alyson K. Fletcher and Philip Schniter and Ulugbek Kamilov},
  journal= {arXiv preprint arXiv:1501.01797},
  year   = {2016}
}