Automatically optimizing heuristics for robust scale-free network design via large language models.

Area of Science:

• Network Science
• Artificial Intelligence
• Computational Complexity

Background:

• Generating scale-free robust networks is challenging due to NP-hard complexity and high-dimensional solution spaces.
• Existing methods often require extensive manual design, trial-and-error, and large labeled datasets.
• Current approaches lack flexibility and adaptability in creating complex network structures.

Purpose of the Study:

• To propose AutoRNet, a novel framework for automated heuristic generation in robust network design.
• To integrate large language models (LLMs) with evolutionary algorithms for efficient network optimization.
• To overcome limitations of current methods by reducing manual intervention and improving adaptability.

Main Methods:

• AutoRNet integrates LLMs with evolutionary algorithms, guided by expert-crafted Network Optimization Strategies (NOSs).
• NOS-based variation operations provide domain-specific prompts to LLMs, incorporating expert knowledge.
• An adaptive fitness function manages degree distribution constraints, balancing convergence and diversity.

Main Results:

• Networks generated by AutoRNet heuristics demonstrate superior robustness compared to existing methods.
• The framework effectively reduces the need for manual heuristic design through structured domain knowledge.
• AutoRNet offers a flexible and adaptive solution for generating robust scale-free network structures.

Conclusions:

• AutoRNet provides an effective and automated approach to designing robust scale-free networks.
• The integration of LLMs and evolutionary algorithms with expert knowledge offers significant advantages.
• The framework's adaptability and reduced manual effort represent a substantial advancement in network optimization.