关键词: 零信任, 软件定义边界, 联邦学习, 去中心化, 信任评估

Abstract: With the increasing blurring of network boundaries, zero trust has emerged as a new paradigm for network security defense. A federated learningbased SDP trust evaluation model and its deployment method are proposed to address the issues of low trust evaluation efficiency and difficulty in effectively protecting user data privacy in the face of massive contextual information and diverse terminal scenarios brought by the zero trust security architecture in the era of big data. This model adopts a decentralized approach to train a global model without sharing raw data, protecting the user data privacy of each distributed SDP controller node. Through experiments and comparative analysis, it has been proven that this zero trust evaluation model can effectively classify malicious and legitimate data streams, and its efficiency is superior to similar literature schemes.

Key words: zero trust, SDP, federated learning, decentralization, trust evaluation