【课题推荐发表期刊】

【课题背景】

Federated Learning (FL) is a collaborative machine learning approach for training models without sharing private data. As FL enables the use of large amounts of data and it has the aim to preserve user privacy, it has become a popular potential solution in many areas, such as cyberattack detection, vehicle networks, smart healthcare, and the Internet of Things (IoT). FL works by leveraging edge computing platforms. The FL approach presents new vulnerabilities and security challenges. The FL suffers attacks against collaborative training, which can be normally categorized as model performance attacks and data privacy attacks. The model performance attacks aim to directly or indirectly affect the performance of the global model, while the data privacy attacks intend to pilfer the data stored on participants. There are already many solutions targeting the two major threats to the FL. However, one possible source of attacks is the communication channel, which has been overlooked by major research work on FL security.

联邦学习（FL）是一种协作式机器学习方法，用于在不共享私有数据的情况下训练模型。由于FL能够使用大量数据，并且旨在保护用户隐私，它已成为许多领域的热门潜在解决方案，例如网络攻击检测、车辆网络、智能医疗和物联网（IoT）。FL通过利用边缘计算平台来实现。然而，FL方法带来了新的脆弱性和安全挑战。FL面临针对协作训练的攻击，这些攻击通常可以分为模型性能攻击和数据隐私攻击。模型性能攻击旨在直接或间接影响全局模型的性能，而数据隐私攻击则意图窃取参与者存储的数据。已经有许多解决方案针对FL的两大主要威胁。然而，在主要的FL安全研究中，忽视了一个可能的攻击来源，就是通信渠道。

本课题将围绕通讯渠道这个攻击来源，开展相应的通信功能、通讯协议等有效的安全解决方案研究，保护通讯渠道免受恶意攻击。

【课题方向参考】

• Design of an efficient communication protocol, that can support the information exchanged between participants and aggregators efficiently with fewer communication resources. 设计一种高效的通信协议，能够以较少的通信资源高效支持参与者和聚合器之间的信息交换

• Optimization of efficient communication functions with heterogeneous device supports. 优化高效通信功能和异构设备支持的功能

• Design of effective security solutions to protect the communication channel against malicious attacks. 设计有效的安全解决方案，保护通信渠道免受恶意攻击

【适合人群】

具有计算机网络，无线网络，信息安全基础知识，逻辑思路清晰，具有较强逻辑推理能力以及较强编程能力。

【课题收获】

• 高质量论文一篇（SCI/CCF定向期刊）

• SCI/CCF期刊投递与发表指导

• 结业证书

【导师介绍】

马教授，QS世界排名前200高校教授，QS排名前50名高校博士毕业

• 长期担任多个SCI期刊和会议的编委或审稿人，包括Computer and Communication Engineering、IEEE Communications Surveys & Tutorials、Wireless Communications and Mobile Computing、Communication Systems等；

• 研究方向：计算机网络，网络安全与AI应用；

• 已发表530多篇国际学术文章，包括260多篇学术期刊文章和260多篇学术会议文章； 6部学术著作的主编和28部学术著作中28个章节的作者；根据谷歌学术统计，所有出版物被引用接近12000 次；H值53。

• 具备丰富的英文写作经验和论文指导经验；掌握Matlab、BAN Logic、and Scyther Tool等软件工具。

【课题安排】

研究周期预估六个月左右，具体视学员情况调整

【推荐阅读】

• 2021 Achieving security and privacy in federated learning systems: Survey, research challenges and future directions