Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Designing Network Algorithms via Large Language Models

Zhiyuan He, Aashish Gottipati, Lili Qiu, Xufang Luo, Kenuo Xu, Yuqing Yang, Francis Y. Yan

TL;DR

Nada is introduced, the first framework to autonomously design network algorithms by leveraging the generative capabilities of large language models by efficiently identifies the top-performing designs through a series of filtering techniques, minimizing the need for full-scale evaluations and significantly reducing computational costs.

Abstract

We introduce NADA, the first framework to autonomously design network algorithms by leveraging the generative capabilities of large language models (LLMs). Starting with an existing algorithm implementation, NADA enables LLMs to create a wide variety of alternative designs in the form of code blocks. It then efficiently identifies the top-performing designs through a series of filtering techniques, minimizing the need for full-scale evaluations and significantly reducing computational costs. Using adaptive bitrate (ABR) streaming as a case study, we demonstrate that NADA produces novel ABR algorithms -- previously unknown to human developers -- that consistently outperform the original algorithm in diverse network environments, including broadband, satellite, 4G, and 5G.

Designing Network Algorithms via Large Language Models

TL;DR

Nada is introduced, the first framework to autonomously design network algorithms by leveraging the generative capabilities of large language models by efficiently identifies the top-performing designs through a series of filtering techniques, minimizing the need for full-scale evaluations and significantly reducing computational costs.

Abstract

We introduce NADA, the first framework to autonomously design network algorithms by leveraging the generative capabilities of large language models (LLMs). Starting with an existing algorithm implementation, NADA enables LLMs to create a wide variety of alternative designs in the form of code blocks. It then efficiently identifies the top-performing designs through a series of filtering techniques, minimizing the need for full-scale evaluations and significantly reducing computational costs. Using adaptive bitrate (ABR) streaming as a case study, we demonstrate that NADA produces novel ABR algorithms -- previously unknown to human developers -- that consistently outperform the original algorithm in diverse network environments, including broadband, satellite, 4G, and 5G.
Paper Structure (12 sections, 5 figures, 5 tables)

This paper contains 12 sections, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Our Approach
  3. Generating Diverse Designs with LLMs
  4. Filtering and Evaluating Designs
  5. Evaluation
  6. Experiment Settings
  7. Designing States
  8. Designing Neural Networks
  9. Early Stopping Mechanism
  10. Insights from Generated Designs
  11. Discussion
  12. Conclusion

Figures (5)

  • Figure 1: Nada workflow. It leverages LLMs to generate a wide range of alternative designs for a network algorithm and employs a series of filtering techniques to efficiently select the most promising designs for further evaluation.
  • Figure 2: The original algorithm design of Pensieve mao2017neural.
  • Figure 3: Test performance of the best states generated by Nada using GPT-3.5 and GPT-4, compared with the original state design throughout the training process. Nada consistently produces state representations that outperform the original design across four network trace sets in simulation.
  • Figure 4: Test performance of the best generated neural network architectures vs. the original in simulation.
  • Figure 5: Comparison between different early stopping classifiers. False and true negative rates are defined in §\ref{['sec:early-stopping-eval']}.