Abstract:Achieving robust networks is a challenging problem due to its NP-hard nature and complex solution space. Current methods, from handcrafted feature extraction to deep learning, have made progress but remain rigid, requiring manual design and large labeled datasets. To address these issues, we propose AutoRNet, a framework that integrates large language models (LLMs) with evolutionary algorithms to generate heuristics for robust network design. We design network optimization strategies to provide domain-specific prompts for LLMs, utilizing domain knowledge to generate advanced heuristics. Additionally, we introduce an adaptive fitness function to balance convergence and diversity while maintaining degree distributions. AutoRNet is evaluated on sparse and dense scale-free networks, outperforming current methods by reducing the need for manual design and large datasets.

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve This paper aims to address the challenges in network robustness design. Specifically, current methods have the following shortcomings when dealing with this issue: 1. **NP-hard Nature**: The optimization problem of network robustness is NP-hard, which means finding the optimal solution is very difficult. 2. **High-dimensional Solution Space**: The network structure is complex and high-dimensional, making the solution space extremely large. 3. **Limitations of Existing Methods**: Existing methods, including manual feature extraction and deep learning approaches, although making some progress, are still rigid and complex, often requiring manual design, trial and error, and a large amount of labeled data. To overcome these challenges, the paper proposes **AutoRNet**, a new framework that combines large language models (LLMs) and evolutionary algorithms to automatically generate complete heuristic methods for network robustness design. By leveraging the intrinsic properties of network structures, effective network optimization strategies (NOS) are designed and combined with LLMs to generate diverse heuristic methods. Additionally, an adaptive fitness function (AFF) is designed to balance convergence and diversity, thereby discovering better heuristic methods. ### Main Contributions 1. **Hybrid Framework**: Developed a framework for iterative collaboration between evolutionary algorithms (EAs) and large language models (LLMs) to generate and optimize heuristic methods for network robustness. 2. **NOS-based Mutation Operations**: Designed mutation operations specific to network problems that can generate problem-specific prompts to guide LLMs in divergent thinking. These NOS are also applicable to other network-related challenges. 3. **Adaptive Fitness Function**: Designed an adaptive fitness function for network problems that transforms hard constraints into soft constraints, enhancing the diversity and robustness of heuristics. 4. **Experimental Validation**: Evaluated the heuristic methods generated by AutoRNet on eight scale-free networks of different sizes and densities, as well as a real-world network, showing that their performance surpasses existing methods. ### Experimental Results The experimental results show that the heuristic methods generated by AutoRNet perform excellently on multiple test graphs, especially in terms of network robustness. Specifically: - **Heuristic-v1**: By combining edge swapping and simulated annealing (SA) strategies while maintaining degree distribution, significantly improved network robustness. - **Heuristic-v2**: By reallocating edges between high-degree and low-degree nodes, significantly improved network robustness. - **Heuristic-v3**: By reallocating edges among the neighbors of high-degree nodes, improved the robustness of the network. These heuristic methods not only outperform or are comparable to manually designed algorithms in performance but also explore new strategies with the same actual cost, demonstrating the potential of AutoRNet in optimizing complex networks.

AutoRNet: Automatically Optimizing Heuristics for Robust Network Design via Large Language Models

Robust Network Enhancement from Flawed Networks.

NetLLM: Adapting Large Language Models for Networking

Designing Network Algorithms via Large Language Models

Large Language Models for Networking: Applications, Enabling Techniques, and Challenges

AutoNE

Large Language Models for Knowledge-Free Network Management: Feasibility Study and Opportunities

Large Language Model (LLM)-enabled Graphs in Dynamic Networking

Machine Learning for Robust Network Design: A New Perspective

Identify Critical Nodes in Complex Network with Large Language Models

Large Language Models Meet Next-Generation Networking Technologies: A Review

Surrogate-Assisted Robust Optimization of Large-Scale Networks Based on Graph Embedding

Large Language Models for Networking: Workflow, Advances and Challenges

Large Language Model Assisted Adversarial Robustness Neural Architecture Search

A Computationally Efficient Evolutionary Algorithm for Multiobjective Network Robustness Optimization

Autonomous Multi-Objective Optimization Using Large Language Model

Deep Insights into Automated Optimization with Large Language Models and Evolutionary Algorithms

Graph Reinforcement Learning for Network Control via Bi-Level Optimization

L-AutoDA: Leveraging Large Language Models for Automated Decision-based Adversarial Attacks

Constructing Robust Cooperative Networks using a Multi-Objective Evolutionary Algorithm

Mixture of Experts for Network Optimization: A Large Language Model-enabled Approach