FedCLF -- Towards Efficient Participant Selection for Federated Learning in Heterogeneous IoV Networks

TL;DR

This work targets Federated Learning in highly dynamic IoV networks where data and devices are heterogeneous and time-critical. It introduces FedCLF, which couples a calibrated loss-based utility for participant selection with a feedback control loop that adjusts client sampling frequency based on end-of-round accuracy. The key contributions include the calibrated utility $U_{FedCLF}$, a dynamic sampling mechanism, and comprehensive evaluation showing up to 16% accuracy gains under strong heterogeneity and improved resource efficiency. The results suggest FedCLF can enable faster, more accurate FL in resource-constrained IoV environments, though limitations include reliance on static CIFAR-10 data and opportunities to explore dynamic datasets and predictive feedback in future work.

Abstract

Federated Learning (FL) is a distributed machine learning technique that preserves data privacy by sharing only the trained parameters instead of the client data. This makes FL ideal for highly dynamic, heterogeneous, and time-critical applications, in particular, the Internet of Vehicles (IoV) networks. However, FL encounters considerable challenges in such networks owing to the high data and device heterogeneity. To address these challenges, we propose FedCLF, i.e., FL with Calibrated Loss and Feedback control, which introduces calibrated loss as a utility in the participant selection process and a feedback control mechanism to dynamically adjust the sampling frequency of the clients. The envisaged approach (a) enhances the overall model accuracy in case of highly heterogeneous data and (b) optimizes the resource utilization for resource constrained IoV networks, thereby leading to increased efficiency in the FL process. We evaluated FedCLF vis-à-vis baseline models, i.e., FedAvg, Newt, and Oort, using CIFAR-10 dataset with varying data heterogeneity. Our results depict that FedCLF significantly outperforms the baseline models by up to a 16% improvement in high data heterogeneity-related scenarios with improved efficiency via reduced sampling frequency.

Figures (5)

• Figure 1: FedCLF architecture in the context of IoV networks.
• Figure 2: Variance in heterogeneity levels across the developed datasets based on different S values.
• Figure 3: Comparison of the moving average accuracy when using $U_{FedCLF}$ (L New) vis-à-vis $U_{Loss}$ (L) and FedAvg on datasets, IID-E and NIID-S50-E.
• Figure 4: Comparison of the moving average accuracy when using $U_{FedCLF}$ with a feedback control (L New with Feedback) vis-à-vis L New and FedAvg on datasets, IID-E and NIID-S50-E.
• Figure 5: Comparison of moving average accuracy of FedCLF with other state-of-the-art models for different types of datasets.