论文简报
cs.NI 2605.28197v1 值得读

Automated Heuristic Design for Network Operations

Reza Namvar, José Gallego, Jose A. Ayala-Romero, Livia Elena Chatzieleftheriou, Andres Garcia-Saavedra, Albert Banchs, Marco Fiore

发布日期：2026-05-27 09:20 相关性：1.0000 价值：0.7400 分类：cs.NI cs.IT
arXiv 摘要 PDF 原文 网页正文

摘要

Network operation relies on heuristics to solve many tasks rapidly and efficiently across the protocol stack. These heuristics are the result of thorough human-driven design rooted in expert knowledge of the target system and problem. Recently, approaches powered by Artificial Intelligence have shown promising results in devising solutions that outperform long-established heuristics in classical problems. We explore the possibility of applying such Automated Heuristic Design (AHD) frameworks to network environments by (i) discussing the general integration of AHD with network operation and the associated challenges, as well as (ii) proposing a practical implementation of AHD for a specific networking task, i.e., 5G decoding. Initial results show how modern AHD tools can devise heuristics for Low-Density Parity Check decoding on par with state-of-the-art solutions implemented in production systems.

相关性判断

high
相关方向
coding_theory wireless_communications network_optimization
判断依据

ArXiv paper is tagged cs.IT and explicitly studies AHD for 5G NR LDPC decoding, a core coding/communications problem with network-operations context.

价值判断

High relevance to cs.IT through 5G NR LDPC decoding and coding-theory/wireless-communications focus. Clear technical substance: LLM-guided evolutionary search, FunSearch-style architecture, sandboxed evaluation, and scoring design applied to a concrete CNU heuristic task. Practical value is meaningful but not urgent: reported LDPC result is comparable to a production Boxplus-phi baseline, not clearly better.

核心问题与主要方法

核心问题

How to adapt automated heuristic design to network operations and use it to discover effective decoding heuristics for 5G NR LDPC processing

场景：LLM-based automated heuristic design for network functions, with a case study on iterative 5G NR LDPC decoding under simulation-based evaluation

主要方法

Decompose a network function into a fixed system skeleton plus a small evolving atomic function, here the LDPC check-node update inside belief propagation. Use an LLM-guided evolutionary loop inspired by FunSearch: islands store candidate code by score clusters, samplers prompt an LLM for function variants, and sandboxed evaluators score feasible candidates. Design a scalar hierarchical score that makes reliability dominate BER and makes BER dominate iteration count, while still providing dense numerical feedback for evolutionary search. Reduce simulation cost by selecting informative boundary contexts where decoding is neither trivial nor consistently failing, and by using asynchronous parallel evaluator workers. The discovered CNU keeps a tanh/atanh-style structure but changes sign/magnitude handling, product aggregation via log-accumulation, intrinsic edge removal, and numerical stability placement relative to Boxplus-family baselines.

关键贡献与后续阅读

关键贡献

Formulates a networking-specific AHD workflow that explicitly addresses computational intractability, evaluation efficiency, stochasticity, atomic task design, scoring, and simulation-cost trade-offs. Implements a distributed LLM-based AHD system for networking with island populations, score clusters, samplers, Qwen3-backed generation, sandboxed evaluators, HTTP orchestration, and asynchronous parallelism. Applies the framework to 5G NR LDPC decoding by evolving CNU code while preserving the surrounding transport-block processing, VNU logic, and CRC-based early stopping pipeline. Provides a concrete scoring design for decoder heuristic search that orders catastrophic failure, undecoded TB count, BER, and iteration count by priority. Reports an interpretable AHD-discovered CNU whose behavior is statistically on par with Boxplus-phi across the evaluated LDPC contexts, supporting feasibility of automated heuristic discovery in a realistic communications pipeline.

研究启发

Does the open-source implementation include enough reproducibility detail to recreate the 1.3e5-program search, context selection, and evaluator settings? How sensitive are the discovered CNU functions to the chosen single boundary context and the 30-TB evaluation batch? What is the actual runtime, memory footprint, and hardware behavior of the AHD-discovered CNU compared with Boxplus, Boxplus-phi, Min-Sum, and Offset Min-Sum? Can the methodology produce gains on other networking tasks where the atomic function is less cleanly isolated than LDPC CNU?

限制与不确定性

Evidence comes from abstract plus partial section structure, not a full-paper review. Generalization from the LDPC case to broader network operations remains unproven. Initial results may depend heavily on simulation scoring choices and evaluator noise.

参考文献

11 条
  1. [1] A. Agarwal, V. Arun, D. Ray, R. Martins, and S. Seshan (2024-04) Towards provably performant congestion control . In NSDI 24 , pp. 951–978 . External Links: ISBN 978-1-939133-39-7 Cited by: §I .
  2. [2] F. L. et al. LLM4AD: large language model for algorithm design . External Links: Link Cited by: §I .
  3. [3] A. Fox, F. D. Pellegrini, and E. Altman (2025) Constrained multi-agent reinforcement learning for task offloading in wireless edge networks . In NeurIPS 2025 Workshop: AI and ML for Next-Generation Wireless Communications and Networking , Cited by: §I .
  4. [4] B. Guler, G. Geraci, and H. Jafarkhani (2025) Robust channel representation for wireless: a multi-task masked contrastive approach . In NeurIPS 2025 Workshop: AI and ML for Next-Generation Wireless Communications and Networking , Cited by: §I .
  5. [5] J. Hoydis, S. Cammerer, F. A. Aoudia, A. Vem, N. Binder, G. Marcus, and A. Keller (2023) Sionna: an open-source library for next-generation physical layer research . External Links: 2203.11854 , Link Cited by: § III-C .
  6. [6] A. Javid and N. G. Prelcic (2025) Physics-informed neural networks for wireless channel estimation with limited pilot signals . In NeurIPS 2025 Workshop: AI and ML for Next-Generation Wireless Communications and Networking , Cited by: §I .
  7. [7] Y. Jia, Y. Mei, and M. Zhang (2023) Learning heuristics with different representations for stochastic routing . IEEE Transactions on Cybernetics 53 ( 5 ), pp. 3205–3219 . External Links: Document Cited by: §I .
  8. [8] F. Liu, X. Tong, M. Yuan, X. Lin, F. Luo, Z. Wang, Z. Lu, and Q. Zhang (2024) Evolution of heuristics: towards efficient automatic algorithm design using large language model . ICML’24 . Cited by: §I , §I .
  9. [9] B. Romera-Paredes, M. Barekatain, A. Novikov, M. Balog, M. P. Kumar, E. Dupont, F. J. R. Ruiz, J. S. Ellenberg, P. Wang, O. Fawzi, P. Kohli, and A. Fawzi (2024-01) Mathematical discoveries from program search with large language models . Nature 625 ( 7995 ), pp. 468–475 ( en ). Cited by: §I , §I , § III-B .
  10. [10] H. Ye, J. Wang, Z. Cao, F. Berto, C. Hua, H. Kim, J. Park, and G. Song (2024) ReEvo: large language models as hyper-heuristics with reflective evolution . NIPS ’24 . External Links: ISBN 9798331314385 Cited by: §I .
  11. [11] C. Yen, S. Abbasloo, and H. J. Chao (2023) Computers can learn from the heuristic designs and master internet congestion control . In ACM SIGCOMM ’23 , pp. 255–274 . External Links: ISBN 9798400702365 , Document Cited by: §I .

底部评论

0 条根评论，可继续回复叠楼

0/2000