![[2602.16038] Heuristic Search as Language-Guided Program Optimization](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16038] Heuristic Search as Language-Guided Program Optimization
Summary
The paper presents a structured framework for Language-Guided Program Optimization, improving Automated Heuristic Design in combinatorial optimization by decomposing the process into modular stages.
Why It Matters
This research addresses the limitations of existing heuristic design methods that rely heavily on manual adjustments and domain expertise. By proposing a modular approach, it enhances the efficiency and adaptability of heuristic search processes, making it relevant for researchers and practitioners in machine learning and optimization.
Key Takeaways
- Introduces a modular framework for heuristic design in optimization.
- Enhances the iterative refinement process through clear stage separation.
- Demonstrates improved performance across diverse real-world applications.
- Integrates existing heuristic methods into a structured pipeline for better outcomes.
- Addresses the challenges of manual trial-and-error in heuristic design.
Computer Science > Neural and Evolutionary Computing arXiv:2602.16038 (cs) [Submitted on 17 Feb 2026] Title:Heuristic Search as Language-Guided Program Optimization Authors:Mingxin Yu, Ruixiao Yang, Chuchu Fan View a PDF of the paper titled Heuristic Search as Language-Guided Program Optimization, by Mingxin Yu and 2 other authors View PDF Abstract:Large Language Models (LLMs) have advanced Automated Heuristic Design (AHD) in combinatorial optimization (CO) in the past few years. However, existing discovery pipelines often require extensive manual trial-and-error or reliance on domain expertise to adapt to new or complex problems. This stems from tightly coupled internal mechanisms that limit systematic improvement of the LLM-driven design process. To address this challenge, we propose a structured framework for LLM-driven AHD that explicitly decomposes the heuristic discovery process into modular stages: a forward pass for evaluation, a backward pass for analytical feedback, and an update step for program refinement. This separation provides a clear abstraction for iterative refinement and enables principled improvements of individual components. We validate our framework across four diverse real-world CO domains, where it consistently outperforms baselines, achieving up to $0.17$ improvement in QYI on unseen test sets. Finally, we show that several popular AHD methods are restricted instantiations of our framework. By integrating them in our structured pipeline, we can u...
