Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

LLM Agent for Hyper-Parameter Optimization

Created by
  • Haebom

Author

Wanzhe Wang, Jianqiu Peng, Menghao Hu, Weihuang Zhong, Tong Zhang, Shuai Wang, Yixin Zhang, Mingjie Shao, Wanli Ni

Outline

In this paper, we design a large-scale language model (LLM) agent for hyperparameter optimization of Warm-Start Particles Swarm Optimization with Crossover and Mutation (WS-PSO-CM) algorithm for unmanned aerial vehicle (UAV) trajectory and communication. Considering that the existing heuristic-based hyperparameter optimization methods have low level of automation and room for performance improvement, we propose an LLM agent that applies an iterative framework and Model Context Protocol (MCP). The LLM agent is configured through a profile that specifies the boundaries of hyperparameters, task objectives, termination conditions, conservative or aggressive strategies for hyperparameter optimization, and LLM configurations. The LLM agent repeatedly calls the WS-PSO-CM algorithm to perform exploration, and terminates the loop according to the termination condition and returns an optimized set of hyperparameters. Experimental results show that the minimum sum rate achieved using the hyperparameters generated by the LLM agent is much higher than that of the heuristic and random generation methods, demonstrating that the LLM agent with the knowledge of PSO and WS-PSO-CM algorithms is useful for finding high-performance hyperparameters.

Takeaways, Limitations

•
Takeaways:
◦
We present a novel method to automatically optimize the hyperparameters of the WS-PSO-CM algorithm by leveraging LLM.
◦
We present the possibility of improving UAV communication performance by generating hyperparameters that outperform heuristic and random generation methods.
◦
Validation of the efficient hyperparameter exploration and optimization capabilities of LLM agents.
•
Limitations:
◦
The performance of the proposed LLM agent may depend on the LLM and WS-PSO-CM algorithms used.
◦
Additional validation of generalization performance for various UAV communication environments and scenarios is needed.
◦
Lack of analysis of the computational cost and training time of LLM agents.
👍
Made with Slashpage