Massively Parallel Benders Decomposition for Correlation Clustering

TL;DR

This paper introduces a parallel Benders decomposition method for correlation clustering in image segmentation, significantly improving optimization efficiency by leveraging parallel processing and advanced Benders techniques.

Contribution

It presents a novel parallel Benders decomposition formulation for correlation clustering, enabling efficient ILP optimization with cycle inequalities and accelerated convergence.

Findings

Enables massive parallelization of correlation clustering optimization.

Accelerates ILP solving through Magnanti-Wong Benders rows.

Demonstrates promising results in image segmentation tasks.

Abstract

We tackle the problem of graph partitioning for image segmentation using correlation clustering (CC), which we treat as an integer linear program (ILP). We reformulate optimization in the ILP so as to admit efficient optimization via Benders decomposition, a classic technique from operations research. Our Benders decomposition formulation has many subproblems, each associated with a node in the CC instance's graph, which are solved in parallel. Each Benders subproblem enforces the cycle inequalities corresponding to the negative weight edges attached to its corresponding node in the CC instance. We generate Magnanti-Wong Benders rows in addition to standard Benders rows, to accelerate optimization. Our Benders decomposition approach provides a promising new avenue to accelerate optimization for CC, and allows for massive parallelization.