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.

Benchmarking Vector, Graph and Hybrid Retrieval Augmented Generation (RAG) Pipelines for Open Radio Access Networks (ORAN)

Created by
  • Haebom

Author

Sarat Ahmad, Zeinab Nezami, Maryam Hafeez, Syed Ali Raza Zaidi

Outline

This paper focuses on the role of generative AI (GenAI) in autonomous optimization of next-generation wireless networks, and in particular on the generation of xApps and rApps using Large Language Models (LLMs) within the Open RAN (O-RAN) architecture. To address the high cost and resource consumption of conventional LLM fine-tuning approaches, we propose a Retrieval-Augmented Generation (RAG) technique. In particular, we compare and evaluate vector-based RAG, GraphRAG, and Hybrid GraphRAG to measure their performance (fidelity, answer relevance, contextual relevance, and factual accuracy) under various question complexities based on the O-RAN specifications. The experimental results show that GraphRAG and Hybrid GraphRAG outperform conventional vector-based RAG, and in particular, Hybrid GraphRAG improves factual accuracy by 8% and GraphRAG improves contextual relevance by 7%.

Takeaways, Limitations

•
Takeaways:
◦
We experimentally demonstrate that GraphRAG and Hybrid GraphRAG are effective for generating LLM-based xApps and rApps in O-RAN environments.
◦
Hybrid GraphRAG contributes to improving factual accuracy, while GraphRAG contributes to improving contextual relevance.
◦
We suggest that the RAG-based approach can alleviate the cost and resource consumption issues of LLM fine-tuning.
•
Limitations:
◦
This study is an evaluation limited to a specific O-RAN specification. Generalizability to various O-RAN implementations and specifications needs to be verified.
◦
Evaluation criteria are limited. More comprehensive evaluation criteria that take into account various aspects need to be developed.
◦
Lack of application and performance verification in real O-RAN environments. Further research in real system integration and operational environments is required.
👍
Made with Slashpage