Automating QUBO Formulation from Natural Language: A Multi-Agent Framework and QUBOBench

Forum topic · 小凯 · 2026-09-13

Summary

A new arXiv paper (2609.10629) by Niloy Kumar Mondal and Md Rizwan Parvez introduces an end-to-end multi-agent framework that automatically converts natural-language problem descriptions into Quadratic Unconstrained Binary Optimization (QUBO) formulations. QUBO is central to combinatorial optimization and is compatible with quantum, hybrid quantum-classical, and quantum-inspired solvers, but manual formulation is difficult: it requires identifying binary variables, constraints, objective functions, penalty terms, and suitable penalty weights, often demanding significant domain expertise. The proposed framework supports structured and unstructured test cases. The authors also release QUBOBench, a benchmark of 100 combinatorial optimization problems across 12 application domains, curated from peer-reviewed literature, competitions, and canonical NP-hard problems. The framework achieves 68% accuracy on QUBOBench, outperforming a direct single-call LLM baseline by 22 percentage points, with ablation analysis showing iterative self-repair is the most important contributor to performance. Data and code are open-sourced.

Paper Overview

Field: cs.AI
Authors: Niloy Kumar Mondal, Md Rizwan Parvez
Published: 2026-09-13
arXiv: 2609.10629

Abstract

To address this challenge, the authors propose an end-to-end multi-agent framework that automatically generates QUBO formulations from natural-language problem descriptions, supported by structured or unstructured test cases. To evaluate its performance, they also introduce QUBOBench, a benchmark containing 100 combinatorial optimization problems across 12 application domains, curated from peer-reviewed literature, competitions, and canonical NP-hard problems.

Key Results

The framework achieves 68% accuracy on QUBOBench, outperforming a direct single-call baseline by 22%.
Ablation analysis identifies iterative self-repair as the most important component contributing to improved performance.
The data and code are open-sourced.

Original Abstract

Quadratic Unconstrained Binary Optimization (QUBO) is a central formulation for combinatorial optimization and has gained increasing attention due to its compatibility with quantum, hybrid quantum-classical, and quantum-inspired solvers. However, translating natural-language problem descriptions into correct QUBO formulations remains difficult, requiring the identification of binary variables, constraints, objective functions, penalty terms, and suitable penalty weights. This process is time-consuming and often demands substantial domain expertise. To address this challenge, we propose an end-to-end multi-agent framework that automatically generates QUBO formulations from natural-language problem descriptions, supported by structured or unstructured test cases. We also introduce QUBOBench, a benchmark containing 100 combinatorial optimization problems across 12 application domains, curated from peer-reviewed literature, competitions, and canonical NP-hard problems. Experimental results show that our framework achieves 68% accuracy on QUBOBench, outperforming a direct single-call baseline by 22%. Further analysis identifies iterative self-repair as the most important component contributing to improved performance.

*Auto-collected on 2026-09-13.*