【Author】
Lu, Yunlong; Huang, Xiaohong; Zhang, Ke; Maharjan, Sabita; Zhang, Yan
【Abstract】In 5G and beyond networks, the increasing inclusion of heterogeneous smart devices and the rising privacy and security concerns, are two crucial challenges in terms of computation complexity and privacy preservation for Artificial Intelligence (AI)-based solutions. In this regard, federated learning emerges as a new technique, which enlarges the scale of training data, and protects the privacy of user data. The development of edge computing makes it possible to apply federated learning to beyond 5G. However, the security of local parameters, the learning quality, and the varying computing and communication resources, are crucial issues that remain unexplored in federated learning schemes. In this article, we propose a block-chain empowered federated learning framework, and present its potential application scenarios in beyond 5G. We enhance the security and privacy by integrating blockchain into a federated learning scheme for maintaining the trained parameters. In particular, we formulate the resource sharing task as a combinational optimization problem while taking resource consumption and learning quality into account. We design a deep reinforcement learning based algorithm to find an optimal solution to the problem. Numerical results show that the proposed scheme achieves high accuracy and good convergence.
【Keywords】Collaborative work; Blockchain; Training; 5G mobile communication; Servers; Computational modeling; Security
【摘要】在 5G 及以后的网络中，越来越多的异构智能设备以及日益增加的隐私和安全问题是基于人工智能 (AI) 的解决方案在计算复杂性和隐私保护方面的两个关键挑战。在这方面，联邦学习作为一种新技术应运而生，它扩大了训练数据的规模，保护了用户数据的隐私。边缘计算的发展使得将联邦学习应用到 5G 之外成为可能。然而，局部参数的安全性、学习质量以及不同的计算和通信资源，是联邦学习方案中仍未探索的关键问题。在本文中，我们提出了一个区块链赋能的联邦学习框架，并展示了其在 5G 之后的潜在应用场景。我们通过将区块链集成到联邦学习方案中来维护训练参数，从而增强安全性和隐私性。特别是，我们将资源共享任务制定为一个组合优化问题，同时考虑资源消耗和学习质量。我们设计了一种基于深度强化学习的算法来找到问题的最佳解决方案。数值结果表明，该方案具有较高的精度和较好的收敛性。
【关键词】协作工作；区块链；训练; 5G移动通信；服务器；计算建模；安全
评论