OptiHive: Ensemble Selection for Learning-Based Optimization via Statistical Modeling

Maxime Bouscary, Saurabh Amin

Keywords: Combinatorial Optimization, Ensemble Methods, Large Language Model

TL;DR: Statistical ensemble modeling for selecting high-quality solvers from generated candidates.

Abstract: Learning-based solvers have emerged as a promising means of tackling complex optimization problems. However, they remain prone to infeasible or suboptimal solutions, and often rely on iterative refinement procedures that incur significant latency. We introduce OptiHive, a framework that enhances solver-generation pipelines through statistical ensemble modeling. OptiHive generates diverse components (solvers, problem instances, and validation tests) in a single batch and filters out erroneous components to ensure fully interpretable outputs. Taking into account the imperfection of the generated components, we employ a statistical model to infer their true performance, enabling principled uncertainty quantification and solver selection. On tasks ranging from traditional combinatorial optimization problems to challenging variants of the Multi-Depot Vehicle Routing Problem, OptiHive significantly outperforms baselines, increasing the optimality rate from 5\% to 92\% on the most complex problems.

Submission Number: 4