LeFi: Learn to Incentivize Federated Learning in Automotive Edge Computing

TL;DR

This work tackles incentivizing participation in federated learning for automotive edge computing under unknown participant sensitivity and non-IID data. It introduces LeFi, a Learn-to-Incentivize Federated learning framework that uses Gaussian Process Regression to learn CAV sensitivity and gradient-based updates to reward allocations under a fixed budget, with KKT-based data-size decisions. The approach demonstrates faster convergence and higher accuracy than baselines like random search and BARA in CIFAR-10 simulations on an RTX 3080 setup, illustrating scalability to large numbers of vehicles. Overall, LeFi provides a practical, data-efficient mechanism to improve global model accuracy in large-scale vehicular FL with heterogeneous participants.

Abstract

Federated learning (FL) is the promising privacy-preserve approach to continually update the central machine learning (ML) model (e.g., object detectors in edge servers) by aggregating the gradients obtained from local observation data in distributed connected and automated vehicles (CAVs). The incentive mechanism is to incentivize individual selfish CAVs to participate in FL towards the improvement of overall model accuracy. It is, however, challenging to design the incentive mechanism, due to the complex correlation between the overall model accuracy and unknown incentive sensitivity of CAVs, especially under the non-independent and identically distributed (Non-IID) data of individual CAVs. In this paper, we propose a new learn-to-incentivize algorithm to adaptively allocate rewards to individual CAVs under unknown sensitivity functions. First, we gradually learn the unknown sensitivity function of individual CAVs with accumulative observations, by using compute-efficient Gaussian process regression (GPR). Second, we iteratively update the reward allocation to individual CAVs with new sampled gradients, derived from GPR. Third, we project the updated reward allocations to comply with the total budget. We evaluate the performance of extensive simulations, where the simulation parameters are obtained from realistic profiling of the CIFAR-10 dataset and NVIDIA RTX 3080 GPU. The results show that our proposed algorithm substantially outperforms existing solutions, in terms of accuracy, scalability, and adaptability.

Figures (8)

• Figure 1: Computing resources needs in different dataset sizes
• Figure 2: Model accuracy varies in different dataset sizes and Non-IID severity
• Figure 3: Convergence of rewards weights.
• Figure 4: Convergence of data selection.
• Figure 5: Convergence of accuracy.