Built independently by an author, for readers. Read the story and support ChapterPal
ChapterPal Logo
Log inSign up

keyword

LLM-based automatic heuristic design

LLM-based automatic heuristic design is an automated computational framework that utilizes large language models to generate, evaluate, and iteratively optimize heuristic algorithms for solving complex optimization problems without requiring manual algorithm development by human experts. In this process, the language model functions as an algorithmic designer by producing executable code or decision rules tailored to specific problem instances such as combinatorial routing, bin packing, or resource scheduling. Candidate heuristics are automatically tested on benchmark tasks, and their performance metrics guide iterative search, evolutionary computation, or tree search mechanisms that prompt the model to refine and evolve the algorithms into increasingly effective solutions.

1 item

Monte Carlo Tree Search for Comprehensive Exploration in LLM-Based Automatic Heuristic Design

Monte Carlo Tree Search for Comprehensive Exploration in LLM-Based Automatic Heuristic Design

Zhi Zheng, Zhuoliang Xie, Zhenkun Wang, Bryan Hooi

OrganizationsGuangdong Provincial Key Laboratory of Fully Actuated System Control Theory and TechnologyNational University of SingaporeSouthern University of Science and Technology

Why you should read this

Proposes a Monte Carlo Tree Search framework for large language model-driven automatic heuristic design that retains and refines temporarily underperforming algorithms in a tree structure to escape local optima in complex optimization tasks.

Handcrafting heuristics for solving complex optimization tasks (e.g., route planning and task allocation) is a common practice but requires extensive domain knowledge. Recently, Large Language Model (LLM)-based automatic heuristic design (AHD) methods have shown promise in generating high-quality heuristics without manual interventions. Existing LLM-based AHD methods employ a population to maintain a fixed number of top-performing LLM-generated heuristics and introduce evolutionary computation (EC) to iteratively enhance the population. However, these population-based procedures cannot fully develop the potential of each heuristic and are prone to converge into local optima. To more comprehensively explore the space of heuristics, this paper proposes to use Monte Carlo Tree Search (MCTS) for LLM-based heuristic evolution. The proposed MCTS-AHD method organizes all LLM-generated heuristics in a tree structure and can better develop the potential of temporarily underperforming heuristics. In experiments, MCTS-AHD delivers significantly higher-quality heuristics on various complex tasks. Our code is available³.

Added

2026-10-03