[Paper Review] Planar Cycle Covering Graphs

Julian Yarkony, Alexander Ihler+1 more|arXiv (Cornell University)|Apr 6, 2011

Graph Theory and Algorithms15 references3 citations

TL;DR

This paper introduces a novel variational lower-bound for planar binary Markov Random Fields (MRFs) by embedding auxiliary nodes on each face of a planar graph, enabling efficient minimum-weight perfect matching to compute the ground state. The method achieves tighter bounds than existing cycle-enforcing approaches and converges 10–100x faster than state-of-the-art algorithms like MPLP and RPM, while provably capturing all cycle constraints via a planar cycle covering (PCC) construction.

ABSTRACT

We describe a new variational lower-bound on the minimum energy configuration of a planar binary Markov Random Field (MRF). Our method is based on adding auxiliary nodes to every face of a planar embedding of the graph in order to capture the effect of unary potentials. A ground state of the resulting approximation can be computed efficiently by reduction to minimum-weight perfect matching. We show that optimization of variational parameters achieves the same lower-bound as dual-decomposition into the set of all cycles of the original graph. We demonstrate that our variational optimization converges quickly and provides high-quality solutions to hard combinatorial problems 10-100x faster than competing algorithms that optimize the same bound.

Motivation & Objective

• To develop a tighter and more efficient variational lower-bound for minimum energy inference in planar binary MRFs.
• To address the computational infeasibility of enforcing constraints over all cycles via full triangulation or iterative cycle addition.
• To provide a unified framework that subsumes tree-reweighted and outer-planar decompositions by capturing all cycles in a single, efficient construction.
• To achieve fast convergence and high-quality solutions by embedding auxiliary nodes per face, enabling reduction to minimum-weight perfect matching.

Proposed method

• The method introduces auxiliary nodes on each face of a planar MRF's embedding to model unary potentials, transforming the original MRF into a form amenable to perfect matching.
• It constructs a new graph representation—Planar Cycle Covering (PCC)—where each face contributes an auxiliary node, enabling exact ground state computation via minimum-weight perfect matching.
• The PCC construction implicitly enforces constraints over all cycles in the original graph, achieving the same lower-bound as dual-decomposition over all cycles.
• The variational parameters are optimized using subgradient methods with Polyak’s step size rule, leveraging incremental updates in Blossom V for efficiency.
• The approach is shown to be equivalent to dual-decomposition over all cycles, providing the tightest possible variational bound for planar binary MRFs.
• The method is evaluated against MPLP and RPM, using Ising grid problems with varying difficulty to compare convergence speed and duality gap.

Experimental results

Research questions

• RQ1 Can a variational bound be constructed that captures all cycle constraints in a planar binary MRF with only constant-factor overhead?
• RQ2 How does the performance of a cycle-covering variational bound compare to iterative cycle-addition methods like MPLP in terms of convergence speed and solution quality?
• RQ3 Can the PCC construction achieve the same lower-bound as dual-decomposition over all cycles while remaining computationally efficient?
• RQ4 Does the PCC method converge faster and with fewer iterations than existing cycle-enforcing algorithms like RPM and MPLP?
• RQ5 Can the PCC framework be extended to non-planar graphs via triangulation and planar embedding?

Key findings

• The PCC method converges 10–100x faster than MPLP and RPM on hard combinatorial problems, with significant speedups observed across all problem sizes.
• For all problem categories (easy, medium, hard), PCC achieves the optimal solution with zero duality gap, while MPLP fails to converge to optimality on the hardest instances.
• PCC successfully solves a higher fraction of problem instances than MPLP and RPM, particularly on larger and more difficult grids.
• The geometric mean convergence time of PCC is substantially lower than both RPM and MPLP across all problem sizes, with consistent performance gains.
• The PCC construction achieves the same theoretical lower-bound as dual-decomposition over all cycles, proving it captures the tightest possible variational bound for planar binary MRFs.
• The incremental optimization using Blossom V and Polyak’s step size rule enables rapid convergence, with minimal overhead from parameter updates.