Many algorithms related to vehicular applications, such as enhanced perception of the environment, benefit from frequent updates and the use of data from multiple vehicles. Federated learning is a promising method to improve the accuracy of algorithms in the context of vehicular networks. However, limited communication bandwidth, varying wireless channel quality, and potential latency requirements may impact the number of vehicles selected for training per communication round and their assigned radio resources. In this work, we characterize the vehicles participating in federated learning based on their importance to the learning process and their use of wireless resources. We then address the joint vehicle selection and resource allocation problem, considering multi-cell networks with multi-user multiple-input multiple-output (MU-MIMO)-capable base stations and vehicles. We propose a “vehicle-beam-iterative” algorithm to approximate the solution to the resulting optimization problem. We then evaluate its performance through extensive simulations, using realistic road and mobility models, for the task of object classification of European traffic signs. Our results indicate that MU-MIMO improves the convergence time of the global model. Moreover, the application-specific accuracy targets are reached faster in scenarios where the vehicles have the same training data set sizes than in scenarios where the data set sizes differ.

Federated learning is an effective method to train a machine learning model without requiring to aggregate the potentially sensitive data of agents in a central server. However, the limited communication bandwidth, the hardware of the agents and a potential application-specific latency requirement impact how many and which agents can participate in the learning process at each communication round. In this paper, we propose a selection metric characterizing each agent’s importance with respect to both the learning process and the resource efficiency of its wireless communication channel. Leveraging this importance metric, we formulate a general agent selection optimization problem, which can be adapted to different environments with latency or resource-oriented constraints. Considering an example wireless environment with latency constraints, the agent selection problem reduces to the 0/1 Knapsack problem, which we solve with a fully polynomial approximation. We then evaluate the agent selection policy in different scenarios, using extensive simulations for an example task of object classification of European traffic signs. The results indicate that agent selection policies which consider both learning and channel aspects provide benefits in terms of the attainable global model accuracy and/or the time needed to achieve a targeted accuracy level. However, in scenarios where agents have a limited number of data samples or where the latency requirement is very stringent, a pure learning-based agent selection policy is shown to be more beneficial during the early or late stages of the learning process.