Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Autonomous Multi-Objective Optimization Using Large Language Model

Yuxiao Huang, Shenghao Wu, Wenjie Zhang, Jibin Wu, Liang Feng, Kay Chen Tan

TL;DR

This work tackles the challenge of designing effective evolutionary operators for multi-objective optimization (MOPs) without expert intervention. It introduces LLMOPT, a framework that autonomously generates, tests, and refines EA operators via Operator Initialization, Dynamic Operator Selection, Operator Crossover, Operator Mutation, Pilot Run & Repair, and Parallel Operator Evaluation. The approach is evaluated on continuous and combinatorial MOPs—CMOPs, MOKPs, and MOTSPs—using multiple baselines and several LLMs, showing competitive or superior performance and highlighting robustness through error-driven dialogue. The results indicate that automated, textually guided operator design can accelerate the development of adaptive optimization strategies, with implications for scalable, autonomous optimization toolchains.

Abstract

Multi-objective optimization problems (MOPs) are ubiquitous in real-world applications, presenting a complex challenge of balancing multiple conflicting objectives. Traditional evolutionary algorithms (EAs), though effective, often rely on domain-specific expertise and iterative fine-tuning, hindering adaptability to unseen MOPs. In recent years, the advent of Large Language Models (LLMs) has revolutionized software engineering by enabling the autonomous generation and refinement of programs. Leveraging this breakthrough, we propose a new LLM-based framework that autonomously designs EA operators for solving MOPs. The proposed framework includes a robust testing module to refine the generated EA operator through error-driven dialogue with LLMs, a dynamic selection strategy along with informative prompting-based crossover and mutation to fit textual optimization pipeline. Our approach facilitates the design of EA operators without the extensive demands for expert intervention, thereby speeding up the innovation of EA operators. Empirical studies across various MOP categories validate the robustness and superior performance of our proposed framework.

Autonomous Multi-Objective Optimization Using Large Language Model

TL;DR

This work tackles the challenge of designing effective evolutionary operators for multi-objective optimization (MOPs) without expert intervention. It introduces LLMOPT, a framework that autonomously generates, tests, and refines EA operators via Operator Initialization, Dynamic Operator Selection, Operator Crossover, Operator Mutation, Pilot Run & Repair, and Parallel Operator Evaluation. The approach is evaluated on continuous and combinatorial MOPs—CMOPs, MOKPs, and MOTSPs—using multiple baselines and several LLMs, showing competitive or superior performance and highlighting robustness through error-driven dialogue. The results indicate that automated, textually guided operator design can accelerate the development of adaptive optimization strategies, with implications for scalable, autonomous optimization toolchains.

Abstract

Multi-objective optimization problems (MOPs) are ubiquitous in real-world applications, presenting a complex challenge of balancing multiple conflicting objectives. Traditional evolutionary algorithms (EAs), though effective, often rely on domain-specific expertise and iterative fine-tuning, hindering adaptability to unseen MOPs. In recent years, the advent of Large Language Models (LLMs) has revolutionized software engineering by enabling the autonomous generation and refinement of programs. Leveraging this breakthrough, we propose a new LLM-based framework that autonomously designs EA operators for solving MOPs. The proposed framework includes a robust testing module to refine the generated EA operator through error-driven dialogue with LLMs, a dynamic selection strategy along with informative prompting-based crossover and mutation to fit textual optimization pipeline. Our approach facilitates the design of EA operators without the extensive demands for expert intervention, thereby speeding up the innovation of EA operators. Empirical studies across various MOP categories validate the robustness and superior performance of our proposed framework.
Paper Structure (26 sections, 5 equations, 11 figures, 6 tables, 5 algorithms)

This paper contains 26 sections, 5 equations, 11 figures, 6 tables, 5 algorithms.

Table of Contents

  1. Introduction
  2. Background
  3. Review of LLM-assisted Program Evolution
  4. Continuous Multi-objective Optimization Problems
  5. Combinatorial Multi-objective Optimization Problems
  6. Multi-objective Knapsack Problems
  7. Multi-objective Traveling Salesman Problems
  8. Proposed Method
  9. Operator Initialization
  10. Dynamic Operator Selection via Scores
  11. Operator Crossover
  12. Operator Mutation
  13. Pilot Run & Repair
  14. Parallel Operator Evaluation
  15. Experimental Study
  16. ...and 11 more sections

Figures (11)

  • Figure 1: Illustration of the proposed multi-objective algorithm evolution via LLM.
  • Figure 2: Prompt design for 'Operator Initialization'
  • Figure 3: Prompt design for 'Operator Crossover'
  • Figure 4: Prompt design for 'Operator Mutation'
  • Figure 5: Prompt design for repairing operators
  • ...and 6 more figures