AutoPBO uses large language models to tune local search solvers for pseudo-Boolean optimization
A research team reports an LLM-powered framework that automates the design of heuristics in local search PBO solvers, an area that previously relied on expert manual tuning.
In plain words
Pseudo-Boolean optimization is a way of modeling combinatorial problems with constraints, and local search solvers that tackle it depend on internal heuristics — rules that steer the search. Designing those heuristics usually takes expert effort. AutoPBO applies large language models to generate such improvements automatically.
What was announced
The authors describe AutoPBO as a framework for automatically enhancing local search solvers for pseudo-Boolean optimization (PBO), a framework for modeling combinatorial problems through pseudo-Boolean constraints. They note that local search solvers perform well on PBO but rely on internal heuristics whose design demands significant expert effort and manual tuning, and that LLMs had not previously been applied to optimizing these solvers.
Evaluation covered four public benchmarks, including one real-world set, one from a PB competition, an integer linear programming optimization benchmark, and a crafted combinatorial benchmark. AutoPBO was compared against six state-of-the-art competitors: local search solvers NuPBO and OraSLS, complete PB solvers PBO-IHS and RoundingSat, and mixed integer programming solvers Gurobi and SCIP. According to the paper, AutoPBO improved significantly on earlier local search approaches while staying competitive with the other state-of-the-art systems, which the authors say suggests a promising route to automating local search solver design.
What it does not show
- The evidence is the paper's abstract only; it reports the authors' own summary of results, and no independent confirmation, detailed benchmark numbers, or availability details are provided.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.