联邦学习在无线网络中的应用

Applications of Federated Learning in Wireless Networks

无线网络将通过引入与融合人工智能技术实现网络泛在智能,已经成为普遍共识。传统集中式模型训练一般需要中央实体来负责数据的汇聚和处理工作,然而网络数据的直接共享困难、隐私需求和训练数据的传输成本较高等带来新的挑战。联邦学习作为新兴的分布式人工智能框架,可以在数据本地化前提下进行多方模型训练,成为未来无线网络实现泛在智能的重要解决方案之一。探索联邦学习在未来无线网络中的应用场景,并从通信效率提升、安全与隐私增强、模型个性化与激励机制等方面总结联邦学习与无线系统结合的研究热点,提出未来发展建议。

Aritificial Intelligence(AI)is generally introduced and integrated into wireless networks to realize ubiquitous network intelligence.Traditional centralized model training generally requires a central server to be responsible for data aggregation and processing.However,it brings threats such as network data privacy leakage,and challenges in high transmission costs of training data.As an emerging distributed AI framework,Federated Learning(FL)can perform multi-party model training under the condition of data localization,and becomes one of the important solutions for the realization of ubiquitous intelligence in future wireless networks.The application scenarios of federated learning in future wireless networks are explored,and the research hotspots of the combination of federated learning and wireless systems are summerized from the aspects of communication efficiency improvement,security and privacy enhancement,model personalization and incentive mechanism,and some suggestions for future development are provided.