Differential Privacy-Based Noise Injection for Mitigating Memory Poisoning in Collaborative Agent Frameworks
International Journal of Information and Communication Technology Trends · 15.03.2026 · 6 (1) · 35-41
DOI: 10.66525/ijictt72 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Information and Communication Technology Trends, online erschienen am 15.03.2026. Verfasst von Daniel Thompson und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Direct memory exchange in collaborative environments increases exposure to adversarial manipulation. This study investigates controlled noise injection inspired by differential privacy to buffer the impact of poisoning. Gaussian noise calibrated to sensitivity bounds is added to shared memory updates before propagation. Analytical bounds quantify the trade-off between poisoning attenuation and performance loss. Experiments on a 64-agent cooperative resource allocation task demonstrate that calibrated noise reduces effective poisoning influence by 46.8% under a 20% injection rate. Overall task utility decreased by only 8.9%, remaining within acceptable performance margins. Differential privacy principles provide an unconventional yet effective mitigation strategy against memory poisoning propagation.
Zitieren
APA
Thompson, D., Carter, E., & Martin, S. (2026). Differential Privacy-Based Noise Injection for Mitigating Memory Poisoning in Collaborative Agent Frameworks. International Journal of Information and Communication Technology Trends, 6(1), 35-41. https://doi.org/10.66525/ijictt72
Vancouver
Thompson D, Carter E, Martin S. Differential Privacy-Based Noise Injection for Mitigating Memory Poisoning in Collaborative Agent Frameworks. International Journal of Information and Communication Technology Trends. 2026;6(1):35-41. doi:10.66525/ijictt72
BibTeX
@article{Thompson2026,
author = {Thompson, Daniel and Carter, Emily and Martin, Sophie},
title = {Differential Privacy-Based Noise Injection for Mitigating Memory Poisoning in Collaborative Agent Frameworks},
journal = {International Journal of Information and Communication Technology Trends},
year = {2026},
volume = {6},
number = {1},
pages = {35--41},
doi = {10.66525/ijictt72}
}
EUVIMED