Human Immune System Inspired Security for Federated Learning-Empowered Internet of Things
ACM Transactions on Internet of Things · 31.05.2025 · 6 (2) · 1-18
DOI: 10.1145/3722562 ↗Kurz gefasst
Fachartikel in der Zeitschrift ACM Transactions on Internet of Things, online erschienen am 14.04.2025. Verfasst von Aashma Uprety und 2 weiteren Autor:innen. Die Arbeit wurde bisher 7-mal zitiert (Crossref) und nennt 44 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: DoD Center of Excellence in AI and Machine Learning (CoE-AIML) at Howard University, U.S. Army Research Laboratory and NSF.
Abstract (Originalsprache)
The emergence of the Internet of Things (IoT) has revolutionized service automation, enabling the development of smart applications. However, the vast interconnectivity of IoT devices not only produces large volumes of data but also creates multiple potential attack surfaces. While Machine Learning (ML) offers insights from IoT data, inherent data privacy and security challenges hinder its effective utilization. Federated Learning (FL) offers privacy-preserving ML for distributed edge devices. Nevertheless, the susceptibility to attacks poses a threat to the integrity of IoT data impacting ML for IoT services and applications. To tackle this challenge and identify IoT devices compromised by attacks like label-flipped data, this article introduces an innovative defense mechanism modeled after the human immune system. Analogous to ‘B’ cells, which detect viruses within the human body, the Reinforcement Learning (RL) agent identifies malicious IoT nodes that participate in federated learning enabled IoT. Subsequently, the FL server, similar to ‘T’ cells in Human immune systems eliminate/destroy infected cells, quarantines/discards the malicious IoT nodes (that are FL clients) and their reported parameters. Like ‘B’ cells and ‘T’ cells work together to defend the human body against infections and diseases, RL agent and FL server work together to defend/secure FL enabled IoT from compromised/malicious IoT devices. Specifically, with the help of Deep Reinforcement Learning (DRL), the RL agent continually monitors model updates from participating IoT nodes during training phase to find malicious nodes and then to isolate or remove those malicious nodes (i.e., parameters) at FL server while aggregating parameters for the global model. The effectiveness of the proposed approach is demonstrated through experiments, where RL agent detects malicious/compromised IoT nodes and FL server discards the parameters from such malicious/compromised IoT nodes. We evaluate our proposed approach using numerical results obtained from experiments where we observe that our approach outperforms the existing state-of-the-art approaches in terms of detection rate, error, and accuracy for enhancing IoT security in FL enabled IoT.
Förderung
- DoD Center of Excellence in AI and Machine Learning (CoE-AIML) at Howard University
- U.S. Army Research Laboratory and NSF
Zitieren
APA
Uprety, A., Rawat, D. B., & Sadler, B. (2025). Human Immune System Inspired Security for Federated Learning-Empowered Internet of Things. ACM Transactions on Internet of Things, 6(2), 1-18. https://doi.org/10.1145/3722562
Vancouver
Uprety A, Rawat DB, Sadler B. Human Immune System Inspired Security for Federated Learning-Empowered Internet of Things. ACM Transactions on Internet of Things. 2025;6(2):1-18. doi:10.1145/3722562
BibTeX
@article{Uprety2025,
author = {Uprety, Aashma and Rawat, Danda B. and Sadler, Brian},
title = {Human Immune System Inspired Security for Federated Learning-Empowered Internet of Things},
journal = {ACM Transactions on Internet of Things},
year = {2025},
volume = {6},
number = {2},
pages = {1--18},
doi = {10.1145/3722562}
}
EUVIMED