Cross-Silo Model-Based Secure Federated Transfer Learning for Flow-Based Traffic Classification
Umer Majeed, Sheikh Salman Hassan, Choong Seon Hong
Abstract
Traffic classification is crucial for autonomous network management. Deep learning-based traffic classification methods are in demand because of their ability to accurately classify even encrypted traffic. Federated learning is a way to collaboratively train learning models with privacy-preservation. Transfer learning allows learning models to share knowledge between tasks from different but related domains. Federated Transfer Learning allows collaborative training of privacy-preserving models with knowledge sharing from source to target domains. In this paper, we did secure federated transfer learning for improvising the training-time and accuracy of the targetfederated-model for traffic classification. The target-federatedmodel outperforms the baseline-federated-model trained from scratch. We implemented a simple cross-silo secure aggregation protocol for security.