posted on 2024-10-16, 15:35authored byWeiquan Ni, Shaoliang Zhu, Md Monjurul Karim, Alia AsheralievaAlia Asheralieva, Jiawen Kang, Zehui Xiong, Carsten Maple
In Internet-of-Vehicles (IoV), smart vehicles can efficiently process various sensing data through federated learning (FL) - a privacy-preserving distributed machine learning (ML) approach that allows collaborative development of the shared ML model without any data exchange. However, traditional FL approaches suffer from poor security against the system noise, e.g., due to low-quality trained data, wireless channel errors, and malicious vehicles generating erroneous results, which affects the accuracy of the developed ML model. To address this problem, we propose a novel FL model based on the concept of Lagrange coded computing (LCC) - a coded distributed computing (CDC) scheme that enables enhancing the system security. In particular, we design the first L-CoFL (Lagrange coded FL) model to improve the accuracy of FL computations in the presence of lowquality trained data and wireless channel errors, and guarantee the system security against malicious vehicles. We apply the proposed L-CoFL model to predict the traffic slowness in IoV and verify the superior performance of our model through extensive simulations.
Funding
Academic Centre of Excellence in Cyber Security Research - University of Warwick
Engineering and Physical Sciences Research Council