Research Area:  Machine Learning

Abstract:

Traffic classification is a crucial technology for ensuring Quality of Service (QoS) in network services and network security management. Deep learning has shown great promise in this area, particularly for classifying encrypted traffic. However, the significant number of samples required for training models presents a challenge. In this paper, we propose a multi-task learning and Federated Learning approach for training multi-task models for encrypted traffic classification in a privacy-protected, multi-enterprise setting. Our proposed two-stage federated multi-task learning scheme, pFedDAMT, aims to address data heterogeneity by first obtaining a global multi-task model that performs well for all tasks and then personalizing and fine-tuning the model with each enterprise’s dataset to generate personalized models. Our experiments demonstrate that pFedDAMT improves prediction accuracy by an average of 1.58% compared to other schemes.

Keywords:  

Author(s) Name:  Xueyu Guan, Run Du, Xiaohan Wang & Haipeng Qu

Publisher name:  Springer

DOI:  10.1007/978-3-031-44213-1_22

Volume Information:  Volume 4,(2023)

Paper Link:   https://link.springer.com/chapter/10.1007/978-3-031-44213-1_22