EUVIMEDObjevte EUVIMED
Open access · CC BY 4.0

Federated Memory Aggregation for Poisoning-Resilient Multi-Agent Collaboration

Cheryl Tan, Marcus Teo, Jia Hui Ong

American Journal Of Big Data · 2026-03-15 · 7 (1) · 240-251

DOI: 10.71465/ajbd3658 ↗

At a glance

Journal article in American Journal Of Big Data, published online on 2026-03-15. Written by Cheryl Tan and 2 co-authors. The full text is freely available (open access, CC BY 4.0).

AbstraktBibliografické údajeCiteRelated studies

Abstract (original language)

Direct memory sharing among agents increases vulnerability to adversarial contamination. This work proposes a federated memory aggregation framework in which local agent memories are aggregated through a secure weighted update mechanism rather than direct synchronization. Each agent computes local memory updates and transmits gradient-like summaries to a central aggregation node. A trimmed-mean aggregation rule discards extreme updates likely caused by poisoning. Evaluation on a 64-agent cooperative optimization benchmark demonstrated that under a 25% poisoning rate, baseline direct-sharing systems experienced a 44.2% performance drop, while the federated aggregation framework limited degradation to 13.6%. Communication overhead increased by 9.8%, remaining within acceptable distributed system thresholds. Federated aggregation enhances robustness against poisoning while preserving collaborative performance.

Cite

APA

Tan, C., Teo, M., & Ong, J. H. (2026). Federated Memory Aggregation for Poisoning-Resilient Multi-Agent Collaboration. American Journal Of Big Data, 7(1), 240-251. https://doi.org/10.71465/ajbd3658

Vancouver

Tan C, Teo M, Ong JH. Federated Memory Aggregation for Poisoning-Resilient Multi-Agent Collaboration. American Journal Of Big Data. 2026;7(1):240-251. doi:10.71465/ajbd3658

BibTeX
@article{Tan2026,
  author = {Tan, Cheryl and Teo, Marcus and Ong, Jia Hui},
  title = {Federated Memory Aggregation for Poisoning-Resilient Multi-Agent Collaboration},
  journal = {American Journal Of Big Data},
  year = {2026},
  volume = {7},
  number = {1},
  pages = {240--251},
  doi = {10.71465/ajbd3658}
}
Start for free →