Towards Automated Knowledge Transfer in Evolutionary Multitasking via Large Language Models

2026-04-01 · Source: cs.NE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, extended

Summary

A new framework, the LLM-assisted Optimization Framework (LLMOF), has been developed to autonomously generate effective and efficient Knowledge Transfer Models (KTMs) for various Evolutionary Multi-task Optimization (EMTO) scenarios. This approach leverages large language models (LLMs) to reduce the need for extensive expert knowledge and human intervention in designing these models. LLMOF facilitates the development of innovative transfer models by optimizing for both transfer performance and computational cost. Empirical studies conducted on ten 50-task EMTO benchmarks from the CEC2024 Competition demonstrated that KTMs generated by LLMOF consistently outperform existing hand-crafted knowledge transfer methods, such as Vertical Crossover (VCM) and Solution Mapping (SMM), in terms of both normalized fitness values and running times. For instance, on benchmark 'WCCI1', KTM* achieved a normalized fitness value of 0.19 and a running time of 67.79, significantly better than VCM's 0.81 and SMM's 0.55.

Key takeaway

For research scientists developing advanced optimization algorithms, this work demonstrates that integrating LLMs into the design process can significantly reduce reliance on domain expertise and human effort. You should consider adopting LLM-assisted frameworks for autonomously generating and refining complex components like knowledge transfer models, especially when balancing multiple objectives such as performance and computational cost. This approach offers a robust path to developing superior optimization solutions.

Key insights

LLMs can autonomously design efficient and effective knowledge transfer models for multi-task optimization.

Principles

• Balance transfer effectiveness and computational efficiency.
• Few-shot chain-of-thought improves LLM understanding.

Method

The LLMOF framework uses LLMs with few-shot chain-of-thought prompting for initialization, generation, and mutation of KTMs, evaluating them based on fitness and running time in a multi-objective optimization loop.

In practice

• Use LLMs to automate complex algorithm design.
• Apply multi-objective evaluation for efficiency and effectiveness.
• Employ few-shot CoT for domain-specific LLM guidance.

Topics

• Evolutionary Multi-task Optimization
• Large Language Models
• Knowledge Transfer Models
• Automated Algorithm Design
• Multi-objective Optimization

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• Controlled Self-Evolution for Algorithmic Code Optimization — Controlled Self-Evolution significantly boosts code optimization efficiency by guiding LLM-based iterative refinement.
• Model Merging in the Era of Large Language Models: Methods, Applications, and Future Directions — Model merging offers a computationally efficient way to combine LLM capabilities without extensive retraining.
• Scaling Behaviors of LLM Reinforcement Learning Post-Training: An Empirical Study in Mathematical Reasoning — Larger LLMs demonstrate superior efficiency and performance in RL post-training for mathematical reasoning.
• EvoSelect: Data-Efficient LLM Evolution for Targeted Task Adaptation — EvoSelect efficiently adapts LLMs by selecting diverse, task-aligned synthetic data using optimal transport and diversity regularization.
• Evolutionary Multi-Task Optimization for LLM-Guided Program Discovery — EMO-STA improves LLM-guided program discovery by evolving shared program archives across task families before adapting them to specific tasks.
• Learning, Fast and Slow: Towards LLMs That Adapt Continually [R] — Combining slow parameter updates with fast context optimization enhances LLM adaptation and mitigates forgetting.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.NE updates on arXiv.org.