Value-Aware Collaborative Data Pricing for Federated Learning in Vehicular Networks

Vehicular federated learning (VFL) is a new paradigm that enables the use of data for distributed training under the premise of protecting the privacy of vehicular nodes (VNs). However, due to the heterogeneity of federated learning data, it is a challenge to evaluate the value of data and design an intelligent pricing scheme to effectively motivate the VNs in the vehicular networks (VNets) to complete learning tasks collaboratively. To this end, in this paper, we consider the value of data and propose a value-aware collaborative data pricing scheme for VFL. In the scheme, we first design a data transaction architecture based on the value of data and the cooperation among VNs. Then, by considering the non-independent and identically distributed (non-IID) degree and the age of data (AoD), we develop the data value model to evaluate the quality of data. Next, based on the requirement of the learning task and the data owned by each VN, we formulate the cooperation of the VNs as a coalition game, where the equilibrium of the coalition game is obtained by designing a distributed coalition formation algorithm. The simulation results show that the proposed scheme can lead to higher utility than the traditional methods.