Intrusion Detection Based on Data Privacy in Cloud-Edge Collaborative Computing Using Federated Learning

Abstract

Listen

Translate article

The Internet of Things has been integrated into every aspect of our modern life, intelligent IoT services and applications are booming, and massive amounts of data are generated every day, many of which contain private information. However, due to limited resources and limited computing power, IoT networks are vulnerable to various types of attacks. Therefore, it is crucial to protect the IoT network from adversarial attacks. In today's technology, applying deep learning to classify traffic is a very effective method. It also brings a problem. In a general cloud server architecture, training data needs to be transmitted to the cloud for processing and model training. The massive data transmission overhead will bring delays in transmission and response, as well as privacy leakage issues. Federated learning (FL) based on cloud-edge collaborative networks has received considerable attention, an emerging framework for training deep learning models from decentralized data. The system sends deep learning algorithms to all edges (data sources) at the same time, trains partial models at each edge, and aggregates these partial models into a learned overall model. User information is not uploaded to the cloud during the entire process. This paper adopts the federated learning framework to detect and classify network traffic attacks, and effectively protect user privacy data.