EUVIMEDEUVIMED entdecken
Crossref

SafeSieve: From Heuristics to Experience in Progressive Pruning for LLM-based Multi-Agent Communication

Ruijia Zhang, Xinyan Zhao, Ruixiang Wang, Sigen Chen, Guibin Zhang, An Zhang, Kun Wang, Qingsong Wen

Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (35) · 29892-29900

DOI: 10.1609/aaai.v40i35.40236 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Ruijia Zhang und 7 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

LLM-based multi-agent systems exhibit strong collaborative capabilities but often suffer from redundant communication and excessive token overhead. Existing methods typically enhance efficiency through pretrained GNNs or greedy algorithms, but often isolate pre- and post-task optimization, lacking a unified strategy. To this end, we present SafeSieve, a progressive and adaptive multi-agent pruning algorithm that dynamically refines the inter-agent communication through a novel dual-mechanism. SafeSieve integrates initial LLM-based semantic evaluation with accumulated performance feedback, enabling a smooth transition from heuristic initialization to experience-driven refinement. Unlike existing greedy Top-k pruning methods, SafeSieve employs 0-extension clustering to preserve structurally coherent agent groups while eliminating ineffective links. Experiments across benchmarks (SVAMP, HumanEval, etc.) showcase that SafeSieve achieves 94.01% average accuracy while reducing token usage by 12.4%-27.8%. Results further demonstrate robustness under prompt injection attacks (1.23% average accuracy drop). In heterogeneous settings, SafeSieve reduces deployment costs by 13.3% while maintaining performance. These results establish SafeSieve as an efficient, GPU-free, and scalable framework for practical multi-agent systems. Our code can be found below.

Zitieren

APA

Zhang, R., Zhao, X., Wang, R., Chen, S., Zhang, G., Zhang, A., Wang, K., & Wen, Q. (2026). SafeSieve: From Heuristics to Experience in Progressive Pruning for LLM-based Multi-Agent Communication. Proceedings of the AAAI Conference on Artificial Intelligence, 40(35), 29892-29900. https://doi.org/10.1609/aaai.v40i35.40236

Vancouver

Zhang R, Zhao X, Wang R, Chen S, Zhang G, Zhang A, et al. SafeSieve: From Heuristics to Experience in Progressive Pruning for LLM-based Multi-Agent Communication. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(35):29892-29900. doi:10.1609/aaai.v40i35.40236

BibTeX
@article{Zhang2026,
  author = {Zhang, Ruijia and Zhao, Xinyan and Wang, Ruixiang and Chen, Sigen and Zhang, Guibin and Zhang, An and Wang, Kun and Wen, Qingsong},
  title = {SafeSieve: From Heuristics to Experience in Progressive Pruning for LLM-based Multi-Agent Communication},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  year = {2026},
  volume = {40},
  number = {35},
  pages = {29892--29900},
  doi = {10.1609/aaai.v40i35.40236}
}
Kostenlos starten →