| 摘要: |
| 车联网与联邦学习的深度结合可以实现快速数据感知，训练出高效的AI模型，开发出一系列智能应用。但是，在车联网环境下，联邦学习更容易受到中毒攻击。如何构建适用于车联网的联邦激励框架，增强抵御恶意车辆的中毒攻击是具有挑战的问题。本文研究车联网下联邦学习、联邦激励机制以及中毒攻击抵御方法，介绍3个领域的国内外研究现状与发展趋势，分析车联网下联邦激励框架现有研究存在的问题和面临的挑战，进一步提出两个关键研究方向，即车联网数据异构性对联邦学习性能的影响及优化策略、恶意攻击场景下联邦激励框架的鲁棒性设计。 |
| 关键词: 车联网 投毒攻击 联邦学习 激励框架 人工智能技术 |
| DOI： |
| 投稿时间：2025-03-03修订日期：2025-04-17 |
| 基金项目:国家自然科学基金项目(62062008)、广西重点科技攻关项目(桂科AD20297125)资助 |
|
| Research and Development of Federated Incentive Framework and Strategy for Resisting Poisoning Attacks in the Internet of Vehicles |
|
GE Zhihui1, XU Dong2, LI Taoshen1
|
| (1.Nanning University, Guangxi University;2.Guangxi University) |
| Abstract: |
| The deep integration of the Internet of vehicles (IoV) and federated learning technologies can achieve fast data perception, train efficient artificial intelligence models, and develop a series of intelligent applications. However, in the environment of the IoV, federated learning is more susceptible to poisoning attacks. How to build a federal incentive framework suitable for the IoVs and enhance resistance to poisoning attacks from malicious vehicles is a challenging problem. This paper comprehensively reviews current research progress and development trends across three critical domains: federated learning in IoV, federated incentive mechanisms, and defense against poisoning attacks. It systematically analyzes the existing problems and challenges of federated incentive frameworks in IoV, and proposes two pivotal research directions: 1) Investigating the impact of IoV data heterogeneity on federated learning performance and developing corresponding optimization strategies; 2) Designing robust federated incentive frameworks resilient to malicious attack scenarios. |
| Key words: Internet of vehicles(IoV) poisoning attacks federated learning incentive framework artificial intelligence technology |
