Next-generation wireless networks push security to the edge, where centralized detection struggles with latency, scale, and privacy. We present a decentralized framework that combines federated learning with computer-vision models operating on compact traffic flow matrices. A lightweight CNN classifies flows in real-time, while federated learning trains across distributed nodes without sharing raw data. On a 5G dataset, the centralized model achieves a weighted F1 of 0.9309, while the 5-client federated model closely follows with 0.9287, showing that model averaging preserves performance under decentralization. Moreover, performance scales consistently with the number of clients (one to five).
Decentralized Security in 5G Networks with Federated Learning and Computer Vision Techniques
Nicola Andriolli;
2025-01-01
Abstract
Next-generation wireless networks push security to the edge, where centralized detection struggles with latency, scale, and privacy. We present a decentralized framework that combines federated learning with computer-vision models operating on compact traffic flow matrices. A lightweight CNN classifies flows in real-time, while federated learning trains across distributed nodes without sharing raw data. On a 5G dataset, the centralized model achieves a weighted F1 of 0.9309, while the 5-client federated model closely follows with 0.9287, showing that model averaging preserves performance under decentralization. Moreover, performance scales consistently with the number of clients (one to five).I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
